GetMax AI Lab — process console

Every number on this page came from a command that ran. Generated 2026-08-09 22:32 UTC · ai-lab d0ae2fb · inference on our own L40S GPUs, us-east-2 — no external provider.

Processes live

1 bd.understand_business

Runs logged

Runs succeeded

1275%

Escalated t0→t1

Fields evidence-verified

5/28

Companies profiled

Read this before trusting a score. Our served model scores 10% on GetMax’s 276-line denial gold set — below the 41.7% you get by always guessing the majority class (frontier control: 85%). So this process does not ask the model to judge; it asks it to quote, and every quote is verified against the fetched page in code. A field whose quote cannot be located is nulled. Fewer true facts beat more false ones.

The Process Equation — technique × model, measured

Both sides get scored: the output and the method. This table is what makes “which tool should the agent reach for” a number instead of an opinion.

Technique	Model	Runs	OK	Avg s	Avg tokens	Escalations
t1headless chromium	max-brain	9	9 (100%)	13.8	9,631	9
t0http + bs4 (no model)	max-brain	5	3 (60%)	8.0	6,550	0
autorouter (t0, escalate on fail/thin)	max-brain	2	0 (0%)	1.7	0	0

bd.understand_business — the audit view

Every kept fact carries the verbatim quote it came from. Every dropped fact says what the model claimed and why it was rejected. Mark these right or wrong and they become the gold set.

www.getmaxsolutions.com ICP 20/100 t1 1/7 evidenced

The entity offers enterprise AI enablement services such as custom AI systems, consulting, and co-branded software platforms. These offerings target organizations seeking technology integration rather than delivering direct patient care. Since the service provided is aimed at enabling businesses through AI solutions without involvement in clinical workflows or treatment delivery, it falls outside the scope of healthcare providers treating patients.

Field	Value	Evidence
legal_name	dropped — quote not found in source	claimed: GetMax Solutions
what_they_sell	Enterprise AI enablement services including custom AI systems, consulting, and co-branded software platforms.	“We built AI into our own company. Now we build it into yours. GetMax runs on 6 AI products we built and operate ourselves — a sales calling engine, a meeting assistant, a denied-claims recovery system, and more. We take”
customer_type	dropped — quote not found in source	claimed: Enterprises seeking operational transformation via proprietary AI applications.
size_signals	dropped — quote not found in source	claimed: $30K/seat cohort program shipped; 67,000+ leads managed by AI.
tech_signals	dropped — quote not found in source	claimed: .NET Core backend framework used internally; hosted on AWS cloud environment.

www.hopkinsmedicine.org ICP 85/100 t1 1/7 evidenced

The entity provides direct patient care across multiple domains such as pediatrics, adult specialties, and general health services. This indicates active delivery of clinical services to patients within the United States, placing it firmly among healthcare providers rather than vendors.

Field	Value	Evidence
legal_name	dropped — quote not found in source	claimed: Johns Hopkins Medicine
what_they_sell	Medical care including pediatric, adult specialties, and general health services.	“We can help you make the right call.”
customer_type	dropped — no quote supplied	claimed: Patients seeking medical care
locations	dropped — quote not found in source	claimed: [Raleigh-Durham NC], Florida, Baltimore MD

www.athenahealth.com ICP 25/100 t1 2/7 evidenced

The entity offers an AI-powered suite encompassing EHR, practice management, medical billing, and related services directly to healthcare providers like medical practices and CHCs. Since its target market consists of entities that provide clinical care—such as physicians’ groups and clinics—it qualifies as a 'provider'. However, because it delivers tools used by clinicians during treatment workflows (e.g., EHRs), it functions more as a technology enabler within the delivery chain rather than bei

Field	Value	Evidence
legal_name	dropped — quote not found in source	claimed: athenahealth
what_they_sell	an AI-powered suite including electronic health record (EHR), practice management, medical billing, patient engagement, and payer solutions.	“technology that relieves burdensome work, helps your organization achieve its goals, and frees clinicians to focus more attention on patient care.”
customer_type	medical practices across various specialties such as primary care, orthopedics, women’s health, behavioral health, urgent care, federally qualified health centers (CHCs), enterprise systems.	“Startup Practices Small Medical Practices Medium to Large Medical Practices Enterprise Health Systems Federally Qualified Health Centers Member-centric Care Organizations Behavioral Health Orthopedics Primary Care Urgent”

stripe.com ICP 15/100 t0 1/7 evidenced

The entity offers financial infrastructure such as payment processing, subscription management, card issuance, and embedded finance solutions. These capabilities serve non-healthcare industries like fintech, e-commerce, and general business services. There is no indication that this company operates within healthcare delivery—no patient treatment, clinical service provision, insurance handling, or direct involvement with medical billing workflows.

Field	Value	Evidence
legal_name	dropped — quote not found in source	claimed: Stripe, LLC
what_they_sell	Financial infrastructure enabling businesses to accept payments, manage subscriptions, issue cards, embed financial services, and monetize through various models.	“Financial infrastructure to grow your revenue. Accept payments, offer financial services and implement custom revenue models – from your first transaction to your billionth.”
customer_type	dropped — quote not found in source	claimed: Businesses of all types and scales including startups, e-commerce vendors, SaaS platforms, marketplaces, media outlets, retail chains, professional service firms and large enterpri
size_signals	dropped — quote not found in source	claimed: $1.9 trillion in payments volume processed annually; serving over 200 million active subscriptions managed on Stripe Billing.

Failed runs — kept visible on purpose

URL	Technique	Error
https://stripe.com	t0	expected string or bytes-like object, got 'list'
https://stripe.com	t0	Expecting ',' delimiter: line 33 column 6 (char 1071)
https://www.hopkinsmedicine.org	auto	403 Client Error: Forbidden for url: https://www.hopkinsmedicine.org/
https://www.athenahealth.com	auto	403 Client Error: Forbidden for url: https://www.athenahealth.com/

The OSS lanes — 711 starred + owned repos, classified

Lane A is split by technique, because a DOM driver and a vision model are not substitutes and differ ~100× in cost. Lane B is split by funnel stage. Buckets marked gap are where we own nothing usable.

Lane A — browser automation & computer use

A1 vision-grounded (screenshot -> x,y click) 2 repos 2 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
14,489	mit	web-infra-dev/midscene	star	2026-08-04	AI-powered, vision-driven UI automation for every platform.
0	mit	sparowjake-troll/midscene	OURS	2026-07-24	AI-powered, vision-driven UI automation for every platform.

A2 LLM browser agent (drives a real browser end-to-end) 17 repos 16 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
66,419	mit	Panniantong/Agent-Reach	star	2026-08-04	Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili,
39,912	apache-2.0	vercel-labs/agent-browser	star	2026-08-04	Browser automation CLI for AI agents
14,489	mit	web-infra-dev/midscene	star	2026-08-04	AI-powered, vision-driven UI automation for every platform.
13,367	noassertion no licence	TencentCloud/TencentDB-Agent-Memory	star	2026-08-03	TencentDB Agent Memory is a team-level memory hub for AI Agents ? turning conversations, docs, and code into f
11,078	apache-2.0	bytebot-ai/bytebot	star archived	2025-09-12	Bytebot is a self-hosted AI desktop agent that automates computer tasks through natural language commands, ope
9,843	mit	pinchtab/pinchtab	star	2026-08-03	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
5,737	apache-2.0	Giskard-AI/giskard-oss	star	2026-08-04	? Open-Source Evaluation & Testing library for LLM Agents
2,128	mit	Sahir619/fable-method	star	2026-07-15	The Fable Workflow: how Claude Fable 5 worked, distilled into skills any model can run, with the eval that kee
1,905	mit	aeonfun/opendia	star	2026-08-04	Connect your browser to AI models. Just use Dia on Chrome, Arc or Firefox.
682	mit	langchain-ai/agentevals	star	2026-07-14	Readymade evaluators for agent trajectories
311	mit	0xK3vin/MegaMemory	star	2026-05-03	Persistent project knowledge graph for coding agents. MCP server with semantic search, in-process embeddings,
0	mit	sparowjake-troll/Agent-Reach	OURS	2026-07-03	Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili,
0	mit	sparowjake-troll/agentevals	OURS	2026-07-14	Readymade evaluators for agent trajectories
0	apache-2.0	sparowjake-troll/bytebot	OURS	2025-09-12	Bytebot is a self-hosted AI desktop agent that automates computer tasks through natural language commands, ope
0	mit	sparowjake-troll/opendia	OURS	2026-07-29	Connect your browser to AI models. Just use Dia on Chrome, Arc or Firefox.
0	mit	sparowjake-troll/page-agent	OURS	2026-07-03	JavaScript in-page GUI agent. Control web interfaces with natural language.
0	mit	sparowjake-troll/pinchtab	OURS	2026-07-26	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and

A3 classic driver (deterministic, cheapest, no model) 4 repos 1 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
72,546	bsd-3-clause	D4Vinci/Scrapling	star	2026-07-30	?? An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!
33,438	agpl-3.0 copyleft	lightpanda-io/browser	star	2026-08-04	Lightpanda: the headless browser designed for AI and automation
1,239	noassertion no licence	bug0inc/passmark	star	2026-07-30	The open-source Playwright library for AI browser regression testing with intelligent caching, auto-healing, a
0	noassertion no licence	sparowjake-troll/passmark	OURS	2026-07-09	The open-source Playwright library for AI browser regression testing with intelligent caching, auto-healing, a

A4 scrape / crawl / site -> markdown or HTML 13 repos 9 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
72,546	bsd-3-clause	D4Vinci/Scrapling	star	2026-07-30	?? An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!
66,419	mit	Panniantong/Agent-Reach	star	2026-08-04	Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili,
9,843	mit	pinchtab/pinchtab	star	2026-08-03	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
7,428	noassertion no licence	tiagozip/cap	star	2026-08-01	Free, open-source and self-hosted CAPTCHA alternative to reCAPTCHA. Privacy-first and powered by proof-of-work
3,383	mit	any4ai/AnyCrawl	star	2026-07-31	AnyCrawl ?: A Node.js/TypeScript crawler that turns websites into LLM-ready data and extracts structured SERP
2,903	mit	zcaceres/markdownify-mcp	star	2026-07-30	A Model Context Protocol server for converting almost anything to Markdown
2,009	noassertion no licence	watercrawl/WaterCrawl	star	2026-05-20	Transform Web Content into LLM-Ready Data
1,099	mit	HackUnderway/SearchPhone	star	2026-07-16	Phone number OSINT toolkit with multi-API search (Google, GitHub, Numverify, Reddit, DuckDuckGo) and automatic
498	agpl-3.0 copyleft	franckferman/MetaDetective	star	2026-07-15	Unleash Metadata Intelligence with MetaDetective. Your Assistant Beyond Metagoofil.
381	mit	akvise/trends-checker	star	2026-04-08	Python CLI for Google Trends analysis with advanced rate limiting, cookie auth, and beautiful terminal outp
349	agpl-3.0 copyleft	AeternaLabsHQ/pullmd	star	2026-08-01	Self-hosted URL- and file-to-Markdown service for humans and AI agents - web pages, documents, images, audio,
0	mit	sparowjake-troll/AnyCrawl	OURS	2026-06-29	AnyCrawl ?: A Node.js/TypeScript crawler that turns websites into LLM-ready data and extracts structured SERP
0	bsd-3-clause	sparowjake-troll/Scrapling	OURS	2026-07-04	?? An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!

A5 full desktop / OS control (beyond the browser) 3 repos 3 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
14,489	mit	web-infra-dev/midscene	star	2026-08-04	AI-powered, vision-driven UI automation for every platform.
11,078	apache-2.0	bytebot-ai/bytebot	star archived	2025-09-12	Bytebot is a self-hosted AI desktop agent that automates computer tasks through natural language commands, ope
0	apache-2.0	sparowjake-troll/bytebot	OURS	2025-09-12	Bytebot is a self-hosted AI desktop agent that automates computer tasks through natural language commands, ope

A6 anti-bot / captcha / proxy / stealth (what makes A2-A4 survive in prod) 7 repos 4 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
72,546	bsd-3-clause	D4Vinci/Scrapling	star	2026-07-30	?? An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!
9,843	mit	pinchtab/pinchtab	star	2026-08-03	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
7,428	noassertion no licence	tiagozip/cap	star	2026-08-01	Free, open-source and self-hosted CAPTCHA alternative to reCAPTCHA. Privacy-first and powered by proof-of-work
4,469	gpl-3.0 copyleft	sensepost/gowitness	star	2026-04-22	? gowitness - a golang, web screenshot utility using Chrome Headless
0	noassertion no licence	sparowjake-troll/cap	OURS	2026-06-21	Free, open-source and self-hosted CAPTCHA alternative to reCAPTCHA. Privacy-first and powered by proof-of-work
0	mit	sparowjake-troll/pinchtab	OURS	2026-07-26	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
0	mit	sparowjake-troll/stealth-browser-mcp	OURS	2026-05-24	The only browser automation that bypasses anti-bot systems. AI writes network hooks, clones UIs pixel-perfect

A7 headless-browser infra (run many at once) 3 repos 1 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
33,438	agpl-3.0 copyleft	lightpanda-io/browser	star	2026-08-04	Lightpanda: the headless browser designed for AI and automation
9,843	mit	pinchtab/pinchtab	star	2026-08-03	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
4,469	gpl-3.0 copyleft	sensepost/gowitness	star	2026-04-22	? gowitness - a golang, web screenshot utility using Chrome Headless

Lane B — the sales funnel

B1 BD / ICP / who-is-this-business-for 4 repos 2 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
60,206	agpl-3.0 copyleft	ultralytics/ultralytics	star	2026-08-04	Ultralytics YOLO26, YOLO11, YOLOv8 ? object detection, instance segmentation, semantic segmentation, image cla
6,816	mit	evidence-dev/evidence	star	2026-02-18	Business intelligence as code: build fast, interactive data visualizations in SQL and markdown
381	mit	akvise/trends-checker	star	2026-04-08	Python CLI for Google Trends analysis with advanced rate limiting, cookie auth, and beautiful terminal outp
0	agpl-3.0 copyleft	sparowjake-troll/ultralytics	OURS	2026-07-13	Ultralytics YOLO26, YOLO11, YOLOv8 ? object detection, instance segmentation, semantic segmentation, image cla

B2 lead sourcing / prospecting 14 repos 7 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
16,148	apache-2.0	Robbyant/lingbot-map	star	2026-07-23	A feed-forward 3D foundation model for reconstructing scenes from streaming data
14,724	agpl-3.0 copyleft	midday-ai/midday	star	2026-06-13	Invoicing, Time tracking, File reconciliation, Storage, Financial Overview & your own Assistant made for Freel
1,585	mit	rawfilejson/awesome-osint-arsenal	star	2026-07-21	OSINT & recon toolkit // 100+ tools, one-command installer, SOCMINT, GEOINT, network recon, dark web, forensic
1,485	noassertion no licence	Dyneteq/reconya	star	2026-07-26	Network reconnaissance and asset discovery tool
1,480	none no licence	GONZOsint/geowifi	star	2024-12-21	Search WiFi geolocation data by BSSID and SSID on different public databases.
1,099	mit	HackUnderway/SearchPhone	star	2026-07-16	Phone number OSINT toolkit with multi-API search (Google, GitHub, Numverify, Reddit, DuckDuckGo) and automatic
498	agpl-3.0 copyleft	franckferman/MetaDetective	star	2026-07-15	Unleash Metadata Intelligence with MetaDetective. Your Assistant Beyond Metagoofil.
439	noassertion no licence	7onez/cti-expert	star	2026-08-01	CTI Expert ? Cyber Threat Intelligence & OSINT analysis skill for Claude Code. 67+ commands, 35 techniques, no
212	mit	aws-samples/sample-specship	star	2026-07-13	Spec-driven autonomous engineering workflow for AI coding agents: recon ? plan ? build ? validate ? ship ? wit
0	mit	sparowjake-troll/awesome-osint-arsenal	OURS	2026-07-21	OSINT & recon toolkit // 100+ tools, one-command installer, SOCMINT, GEOINT, network recon, dark web, forensic
0	mit	sparowjake-troll/GenRecon	OURS	2026-06-29	-
0	agpl-3.0 copyleft	sparowjake-troll/midday	OURS	2026-06-13	Invoicing, Time tracking, File reconciliation, Storage, Financial Overview & your own Assistant made for Freel
0	noassertion no licence	sparowjake-troll/reconya	OURS	2026-07-26	Network reconnaissance and asset discovery tool
0	mit	sparowjake-troll/SearchPhone	OURS	2026-07-05	Phone number OSINT toolkit with multi-API search (Google, GitHub, Numverify, Reddit, DuckDuckGo) and automatic

B3 enrich / verify (does the contact actually exist) 2 repos 2 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
1,099	mit	HackUnderway/SearchPhone	star	2026-07-16	Phone number OSINT toolkit with multi-API search (Google, GitHub, Numverify, Reddit, DuckDuckGo) and automatic
0	mit	sparowjake-troll/SearchPhone	OURS	2026-07-05	Phone number OSINT toolkit with multi-API search (Google, GitHub, Numverify, Reddit, DuckDuckGo) and automatic

B4 outreach / sequencing / email + call 18 repos 14 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
18,661	mit	docker-mailserver/docker-mailserver	star	2026-08-03	Production-ready fullstack but simple mail server (SMTP, IMAP, LDAP, Antispam, Antivirus, etc.) running inside
12,412	mit	rmyndharis/OpenWA	star	2026-08-03	Free, Open Source, Self-Hosted WhatsApp API Gateway
11,248	mit	wangrongding/wechat-bot	star	2026-07-08	? Multi-platform IM AI Agent for Telegram, WhatsApp, Lark, and WeChat. Connects ChatGPT / Claude / Kimi / Deep
4,579	agpl-3.0 copyleft	usesend/useSend	star	2026-07-24	Open source alternative to Resend, Sendgrid, Postmark etc.
4,372	mit	aldinokemal/go-whatsapp-web-multidevice	star	2026-08-04	GOWA - WhatsApp REST API with support for UI, Multi Account, Webhooks, and MCP, and Chatwoot. Built with Golan
3,363	mit	typpo/textbelt	star	2024-03-28	Free API for outgoing SMS
3,005	mit	vernu/textbee	star	2026-08-04	open-source sms-gateway. turn any android phone into an sms gateway
1,368	gpl-3.0 copyleft	rafatosta/zapzap	star	2026-08-03	WhatsApp desktop application written in PyQt6 + PyQt6-WebEngine.
87	mit	hoodini/whatsapp-public-logan	star	2026-02-13	Open-source WhatsApp bot with AI capabilities - Logan by YUV.AI
16	noassertion no licence	mylinehub/omnichannel-crm	star	2026-02-07	Open-Source Omnichannel CRM with Premium Autodialer Description: bring all features together?voice, SMS, Whats
1	mit	sparowjake-troll/whatsapp-public-logan	OURS+star	2026-02-13	Open-source WhatsApp bot with AI capabilities - Logan by YUV.AI
1	mit	sparowjake-troll/wechat-bot	OURS+star	2026-07-08	? Multi-platform IM AI Agent for Telegram, WhatsApp, Lark, and WeChat. Connects ChatGPT / Claude / Kimi / Deep
0	mit	sparowjake-troll/docker-mailserver	OURS	2026-07-06	Production-ready fullstack but simple mail server (SMTP, IMAP, LDAP, Antispam, Antivirus, etc.) running inside
0	mit	sparowjake-troll/go-whatsapp-web-multidevice	OURS	2026-07-21	GOWA - WhatsApp REST API with support for UI, Multi Account, Webhooks, and MCP, and Chatwoot. Built with Golan
0	mit	sparowjake-troll/OpenWA	OURS	2026-07-06	Free, Open Source, Self-Hosted WhatsApp API Gateway
0	mit	sparowjake-troll/textbee	OURS	2026-07-10	open-source sms-gateway. turn any android phone into an sms gateway
0	mit	sparowjake-troll/textbelt	OURS	2024-03-28	Free API for outgoing SMS
0	agpl-3.0 copyleft	sparowjake-troll/useSend	OURS	2026-07-13	Open source alternative to Resend, Sendgrid, Postmark etc.

B5 CRM / system of record / pipeline 13 repos 6 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
33,438	agpl-3.0 copyleft	lightpanda-io/browser	star	2026-08-04	Lightpanda: the headless browser designed for AI and automation
20,556	agpl-3.0 copyleft	openobserve/openobserve	star	2026-08-04	Open source observability platform for logs, metrics, traces, frontend monitoring, pipelines and LLM observabi
9,843	mit	pinchtab/pinchtab	star	2026-08-03	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
4,372	mit	aldinokemal/go-whatsapp-web-multidevice	star	2026-08-04	GOWA - WhatsApp REST API with support for UI, Multi Account, Webhooks, and MCP, and Chatwoot. Built with Golan
2,165	mit	rushindrasinha/youtube-shorts-pipeline	star	2026-06-09	Automated YouTube Shorts pipeline: news ? script ? AI visuals ? voiceover ? captions ? upload
1,793	noassertion no licence	YetiForceCompany/YetiForceCRM	star archived	2024-09-13	We've moved! For more information, visit https://github.com/YetiForceCompany/YetiForce
1,122	none no licence	TIGER-AI-Lab/OpenResearcher	star	2026-06-10	OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis
817	apache-2.0	Nike-Inc/koheesio	star	2026-08-01	Python framework for building efficient data pipelines. It promotes modularity and collaboration, enabling the
16	noassertion no licence	mylinehub/omnichannel-crm	star	2026-02-07	Open-Source Omnichannel CRM with Premium Autodialer Description: bring all features together?voice, SMS, Whats
0	mit	sparowjake-troll/go-whatsapp-web-multidevice	OURS	2026-07-21	GOWA - WhatsApp REST API with support for UI, Multi Account, Webhooks, and MCP, and Chatwoot. Built with Golan
0	agpl-3.0 copyleft	sparowjake-troll/openobserve	OURS	2026-07-12	Open source observability platform for logs, metrics, traces, frontend monitoring, pipelines and LLM observabi
0	none no licence	sparowjake-troll/OpenResearcher	OURS	2026-06-10	OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis
0	mit	sparowjake-troll/youtube-shorts-pipeline	OURS	2026-06-09	Automated YouTube Shorts pipeline: news ? script ? AI visuals ? voiceover ? captions ? upload

B6 SEO / content / marketing ops 22 repos 13 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
32,503	noassertion no licence	langfuse/langfuse	star	2026-08-04	? Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datas
20,556	agpl-3.0 copyleft	openobserve/openobserve	star	2026-08-04	Open source observability platform for logs, metrics, traces, frontend monitoring, pipelines and LLM observabi
10,385	mit	every-app/open-seo	star	2026-07-30	Open source alternative to Semrush and Ahrefs
8,093	noassertion no licence	growthbook/growthbook	star	2026-08-04	Open Source Feature Flags, Experimentation, and Product Analytics
6,816	mit	evidence-dev/evidence	star	2026-02-18	Business intelligence as code: build fast, interactive data visualizations in SQL and markdown
3,383	mit	any4ai/AnyCrawl	star	2026-07-31	AnyCrawl ?: A Node.js/TypeScript crawler that turns websites into LLM-ready data and extracts structured SERP
1,770	apache-2.0	apache/ossie	star	2026-07-31	Apache Ossie, industry wide specification effort to standardize how we exchange semantic metadata across analy
713	mit	lyon-industries/graphrag-workbench	star	2026-07-18	Interactive 3D visualization of knowledge graphs generated by Microsoft GraphRAG. Explore entities, relationsh
644	mit	Auriti-Labs/geo-optimizer-skill	star	2026-07-20	Open-source Answer Engine Optimization (AEO) & Generative Engine Optimization (GEO) toolkit ? audit, optimize
498	agpl-3.0 copyleft	franckferman/MetaDetective	star	2026-07-15	Unleash Metadata Intelligence with MetaDetective. Your Assistant Beyond Metagoofil.
494	mit	crawlseo/crawlseo	star	2026-07-29	Open-source SEO monitoring. GSC + Site Crawler + Core Web Vitals in one dashboard.
381	mit	akvise/trends-checker	star	2026-04-08	Python CLI for Google Trends analysis with advanced rate limiting, cookie auth, and beautiful terminal outp
310	none no licence	danilobatson/ai-trading-agent-gemini	star	2026-01-05	AI Trading Agent that transforms social media sentiment into actionable trading signals using LunarCrush analy
16	noassertion no licence	mylinehub/omnichannel-crm	star	2026-02-07	Open-Source Omnichannel CRM with Premium Autodialer Description: bring all features together?voice, SMS, Whats
0	mit	sparowjake-troll/AnyCrawl	OURS	2026-06-29	AnyCrawl ?: A Node.js/TypeScript crawler that turns websites into LLM-ready data and extracts structured SERP
0	mit	sparowjake-troll/crawlseo	OURS	2026-07-21	Open-source SEO monitoring. GSC + Site Crawler + Core Web Vitals in one dashboard.
0	none no licence	sparowjake-troll/getmax-content-pilot	OURS	2026-07-29	-
0	mit	sparowjake-troll/graphrag-workbench	OURS	2025-09-09	Interactive 3D visualization of knowledge graphs generated by Microsoft GraphRAG. Explore entities, relationsh
0	noassertion no licence	sparowjake-troll/growthbook	OURS	2026-07-25	Open Source Feature Flags, Experimentation, and Product Analytics
0	agpl-3.0 copyleft	sparowjake-troll/MetaDetective	OURS	2026-07-04	Unleash Metadata Intelligence with MetaDetective. Your Assistant Beyond Metagoofil.
0	mit	sparowjake-troll/open-seo	OURS	2026-07-16	Open source alternative to Semrush and Ahrefs
0	apache-2.0	sparowjake-troll/ossie	OURS	2026-07-16	Apache Ossie, industry wide specification effort to standardize how we exchange semantic metadata across analy

B7 proposal / quote / contract / billing (close the deal) 46 repos 20 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
126,263	mit	garrytan/gstack	star	2026-07-15	Use Garry Tan's exact Claude Code setup: 23 opinionated tools that serve as CEO, Designer, Eng Manager, Releas
120,479	mit	shadcn-ui/ui	star	2026-08-04	A set of beautifully-designed, accessible components and a code distribution platform. Works with your favorit
94,525	noassertion no licence	iluwatar/java-design-patterns	star	2026-07-31	Design patterns implemented in Java
45,159	noassertion no licence	karanpratapsingh/system-design	star	2026-07-08	Learn how to design systems at scale and prepare for system design interviews
36,332	agpl-3.0 copyleft	nextcloud/server	star	2026-08-04	?? Nextcloud server, a safe home for all your data
35,455	mit	refinedev/refine	star	2026-06-05	A React Framework for building internal tools, admin panels, dashboards & B2B apps with unmatched flexibility
33,438	agpl-3.0 copyleft	lightpanda-io/browser	star	2026-08-04	Lightpanda: the headless browser designed for AI and automation
24,844	mit	emilkowalski/skills	star	2026-08-02	Skills for Designers and Engineers.
13,451	gpl-3.0 copyleft	safing/portmaster	star	2026-06-16	? Love Freedom - ? Block Mass Surveillance
11,267	none no licence	mhadidg/software-architecture-books	star	2023-03-15	A comprehensive list of books on Software Architecture.
6,790	noassertion no licence	superdesigndev/superdesign	star	2026-06-29	AI Product Design Agent - Open Source
6,024	mit	nolly-studio/cult-ui	star	2026-07-22	Components crafted for Design Engineers. Styled using Tailwind CSS, fully compatible with Shadcn, and easy to
5,472	cc0-1.0	PanXProject/awesome-certificates	star	2026-06-30	Curated list of 20,000+ hours and 200+ free courses with certificates in IT, CS, Design and Business.
5,268	mit	PacktPublishing/LLM-Engineers-Handbook	star	2026-04-22	The LLM's practical guide: From the fundamentals to deploying advanced LLM and RAG apps to AWS using LLMOps be
4,703	none no licence	ant-design/x	star	2026-08-04	Craft AI-driven interface effortlessly?
4,082	mit	MengTo/Skills	star	2026-08-01	Agent skills for designers and builders using Codex, Claude, Cursor, and other AI coding agents
3,300	noassertion no licence	DavidHDev/canvas-ui	star	2026-08-03	A library of creative canvas components. Real HTML with WebGL effects running over it. React, Vue, Svelte, van
2,303	none no licence	evoiz/Agentic-Design-Patterns	star	2026-07-24	-
2,174	mit	Paymenter/Paymenter	star	2026-08-04	Free and open-source webshop solution for hostings
1,903	none no licence	Luosunce/material-design-data	star	2023-03-16	?? Material Design ?????????
1,225	noassertion no licence	uvarov-frontend/vanilla-calendar-pro	star	2026-07-01	The Vanilla Calendar Pro is a versatile JavaScript date and time picker component with TypeScript support, mak
1,181	gpl-3.0 copyleft	twelvesec/PwnPad	star	2025-12-05	PwnPad is an affordable, hands-on hardware hacking platform built for practical learning. It features a range
753	mit	nexu-io/codex-slides	star	2026-07-21	? Open-source AI slide studio inside Codex: image-native decks, every slide a full visual canvas. ? 10+ high-q
621	gpl-3.0 copyleft	AHS12/thoth-blueprint	star	2026-06-21	Free visual database design tool with drag-and-move editing, multi-format export, and migration generation for
582	apache-2.0	evan361425/flutter-pos-system	star	2026-08-03	An open-source Flutter POS system designed for small restaurants and businesses.
527	apache-2.0	Saganaki22/ComfyUI-OmniVoice-TTS	star	2026-06-11	OmniVoice TTS nodes for ComfyUI - Zero-shot multilingual text-to-speech with voice cloning, voice design, and
276	noassertion no licence	EA1FUO/AntennaSim	star	2026-07-30	Web-based antenna simulator powered by NEC2. Design, simulate, and analyze antennas with real-time 3D visualiz
265	agpl-3.0 copyleft	HAKORADev/VODER	star	2026-07-29	Voice Operation and Design Engine with Reproduction capabilities
24	mit	fred-terzi/totem-llm	star	2026-07-31	Totem LLM - Your Mind Leads: Private AI agent designed to run on your hardware with your private data leading.
1	mit	sparowjake-troll/gstack	OURS+star	2026-07-15	Use Garry Tan's exact Claude Code setup: 23 opinionated tools that serve as CEO, Designer, Eng Manager, Releas
0	none no licence	sparowjake-troll/Agentic-Design-Patterns	OURS	2026-07-24	-
0	noassertion no licence	sparowjake-troll/AntennaSim	OURS	2026-07-19	Web-based antenna simulator powered by NEC2. Design, simulate, and analyze antennas with real-time 3D visualiz
0	mit	sparowjake-troll/astryx	OURS	2026-06-28	An open source design system that's fully customizable and agent ready
0	agpl-3.0 copyleft	sparowjake-troll/browser	OURS	2026-07-25	Lightpanda: the headless browser designed for AI and automation
0	apache-2.0	sparowjake-troll/ComfyUI-OmniVoice-TTS	OURS	2026-06-11	OmniVoice TTS nodes for ComfyUI - Zero-shot multilingual text-to-speech with voice cloning, voice design, and
0	mit	sparowjake-troll/cult-ui	OURS	2026-05-21	Components crafted for Design Engineers. Styled using Tailwind CSS, fully compatible with Shadcn, and easy to
0	apache-2.0	sparowjake-troll/flutter-pos-system	OURS	2026-07-12	An open-source Flutter POS system designed for small restaurants and businesses.
0	noassertion no licence	sparowjake-troll/java-design-patterns	OURS	2026-07-31	Design patterns implemented in Java
0	noassertion no licence	sparowjake-troll/superdesign	OURS	2026-06-29	AI Product Design Agent - Open Source
0	noassertion no licence	sparowjake-troll/system-design	OURS	2026-07-08	Learn how to design systems at scale and prepare for system design interviews

B8 agent orchestration for a business process 35 repos 25 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
152,840	mit	langflow-ai/langflow	star	2026-08-04	Langflow is a powerful tool for building and deploying AI-powered agents and workflows.
83,112	mit	OpenHands/OpenHands	star	2026-08-04	? OpenHands: AI-Driven Development
45,922	mit	rohitg00/ai-engineering-from-scratch	star	2026-08-04	Learn it. Build it. Ship it for others.
32,503	noassertion no licence	langfuse/langfuse	star	2026-08-04	? Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datas
27,064	apache-2.0	PrefectHQ/fastmcp	star	2026-08-04	? The fast, Pythonic way to build MCP servers and clients.
9,843	mit	pinchtab/pinchtab	star	2026-08-03	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
8,107	apache-2.0	omnigent-ai/omnigent	star	2026-08-04	Omnigent is an open-source AI agent framework and meta-harness: orchestrate Claude Code, Codex, Cursor, Pi, an
7,471	noassertion no licence	refly-ai/refly	star	2026-07-29	The first open-source agent skills builder. Define skills by vibe workflow, run on Claude Code, Cursor, Codex
5,416	agpl-3.0 copyleft	elder-plinius/T3MP3ST	star	2026-08-02	autonomous red teaming platform; multi-agent offensive-security meta-harness
4,061	mit	FareedKhan-dev/all-agentic-architectures	star	2026-06-22	35 production-grade agentic AI architectures (Reflexion, LATS, GraphRAG, MemGPT, Voyager, BrowserAgent, ...) ?
3,942	mpl-2.0	offen/docker-volume-backup	star	2026-08-04	Backup Docker volumes locally or to any S3, WebDAV, Azure Blob Storage, Dropbox, Google Drive or SSH compatibl
3,927	apache-2.0	Osmantic/ODS	star	2026-07-31	Turn your PC, Mac, or Linux box into an AI server. LLM inference, chat UI, voice, agents, workflows, RAG, and
3,800	mit	lemony-ai/cascadeflow	star	2026-07-01	Cascading runtime for AI agents. Optimize cost, latency, quality, and policy decisions inside the agent loop.
1,445	apache-2.0	patoles/agent-flow	star	2026-07-11	Real-time visualization of Claude Code agent orchestration ? see your agents think, branch, and coordinate as
1,292	mit	karam-ajaj/atlas	star	2026-07-05	Open-source tool for network discovery, visualization, and monitoring. Built with Go, FastAPI, and React, supp
1,049	apache-2.0	alookai/alook	star	2026-08-04	The collaboration layer for your AI workforce. Run a team of AI agents that coordinate over email, share memor
890	apache-2.0	swiftwave-org/swiftwave	star	2026-08-04	Self-hosted lightweight PaaS solution to deploy and manage your applications on any VPS [Your own self-hosted
811	mit	SethGammon/Citadel	star	2026-08-04	The operating layer for Claude Code + OpenAI Codex: persistent project memory, intent routing, safety hooks, c
412	none no licence	drivebase/drivebase	star	2026-06-03	Drivebase is a cloud-agnostic file management platform for people and teams using multiple storage providers.
411	mit	sharpdeveye/maestro	star	2026-04-29	Workflow fluency for AI coding agents. 1 core skill ? 25 commands ? 7 domain references ? memory layer ? audit
363	gpl-3.0 copyleft	martin98-afk/CanvasMind	star	2026-07-10	A modern low-code visual programming IDE built on NodeGraphQt and qfluentwidgets, supporting drag-and-drop com
260	apache-2.0	inclusionAI/Avernet	star	2026-08-04	Distributed agent coordination platform where agents live, connect, coordinate, execute, and evolve together.
244	mit	AgriciDaniel/claude-canvas	star	2026-04-10	AI-orchestrated visual production for Obsidian Canvas. Create presentations, flowcharts, mood boards, knowledg
57	apache-2.0	maximem-ai/maximem_synap_sdk	star	2026-08-04	Maximem Synap is the memory layer that makes AI agents remember. 92% LongMemEval, 93.2% on LOCOMO. Works nativ
24	mit	tyhh00/HNA-Code	star	2026-07-27	Mission control for parallel Claude Code & Codex sessions ? a live grid of coding agents that glow when they n
8	none no licence	oss-codes/voice-cost	star	2026-07-24	Open Voice AI cost, SIP, realtime pricing, latency benchmark, country deployment, Asterisk, OSS framework, and
0	apache-2.0	sparowjake-troll/agent-flow	OURS	2026-07-11	Real-time visualization of Claude Code agent orchestration ? see your agents think, branch, and coordinate as
0	gpl-3.0 copyleft	sparowjake-troll/CanvasMind	OURS	2026-07-10	A modern low-code visual programming IDE built on NodeGraphQt and qfluentwidgets, supporting drag-and-drop com
0	mit	sparowjake-troll/claude-canvas	OURS	2026-04-10	AI-orchestrated visual production for Obsidian Canvas. Create presentations, flowcharts, mood boards, knowledg
0	mit	sparowjake-troll/langflow	OURS	2026-07-25	Langflow is a powerful tool for building and deploying AI-powered agents and workflows.
0	apache-2.0	sparowjake-troll/omnigent	OURS	2026-06-28	Omnigent is an open-source AI agent framework and meta-harness: orchestrate Claude Code, Codex, Cursor, Pi, an
0	noassertion no licence	sparowjake-troll/OpenHands	OURS	2026-07-10	? OpenHands: AI-Driven Development
0	mit	sparowjake-troll/pinchtab	OURS	2026-07-26	High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and
0	agpl-3.0 copyleft	sparowjake-troll/T3MP3ST	OURS	2026-07-05	autonomous red teaming platform; multi-agent offensive-security meta-harness
0	none no licence	sparowjake-troll/voice-cost	OURS	2026-07-24	Open Voice AI cost, SIP, realtime pricing, latency benchmark, country deployment, Asterisk, OSS framework, and

B9 eval / A-B / scoring the agents themselves 21 repos 14 permissive

Stars	Licence	Repo	Ours?	Pushed	What it is
46,431	agpl-3.0 copyleft	KeygraphHQ/shannon	star	2026-08-04	Shannon is an AI pentester for web applications and APIs. It analyzes your source code, identifies attack vect
32,503	noassertion no licence	langfuse/langfuse	star	2026-08-04	? Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datas
23,926	mit	promptfoo/promptfoo	star	2026-08-04	Test your prompts, agents, and RAGs. Red teaming/pentesting/vulnerability scanning for AI. Compare performance
20,556	agpl-3.0 copyleft	openobserve/openobserve	star	2026-08-04	Open source observability platform for logs, metrics, traces, frontend monitoring, pipelines and LLM observabi
17,401	apache-2.0	confident-ai/deepeval	star	2026-08-04	The LLM Evaluation Framework
9,953	apache-2.0	Tracer-Cloud/opensre	star	2026-08-04	Build your own AI SRE agents. The open source toolkit for the AI era.
8,690	apache-2.0	NVIDIA/garak	star	2026-08-04	the LLM vulnerability scanner
5,737	apache-2.0	Giskard-AI/giskard-oss	star	2026-08-04	? Open-Source Evaluation & Testing library for LLM Agents
5,416	agpl-3.0 copyleft	elder-plinius/T3MP3ST	star	2026-08-02	autonomous red teaming platform; multi-agent offensive-security meta-harness
2,684	mit	Pouzor/homelable	star	2026-08-01	Self-hosted homelab infrastructure visualizer ? interactive network diagram with live status monitoring
1,150	mit	langchain-ai/openevals	star	2026-08-03	Readymade evaluators for your LLM apps
682	mit	langchain-ai/agentevals	star	2026-07-14	Readymade evaluators for agent trajectories
0	mit	sparowjake-troll/agentevals	OURS	2026-07-14	Readymade evaluators for agent trajectories
0	apache-2.0	sparowjake-troll/deepeval	OURS	2026-07-24	The LLM Evaluation Framework
0	apache-2.0	sparowjake-troll/garak	OURS	2026-07-24	the LLM vulnerability scanner
0	apache-2.0	sparowjake-troll/giskard-oss	OURS	2026-07-16	? Open-Source Evaluation & Testing library for LLM Agents
0	noassertion no licence	sparowjake-troll/langfuse	OURS	2026-07-25	? Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datas
0	mit	sparowjake-troll/openevals	OURS	2026-07-25	Readymade evaluators for your LLM apps
0	agpl-3.0 copyleft	sparowjake-troll/openobserve	OURS	2026-07-12	Open source observability platform for logs, metrics, traces, frontend monitoring, pipelines and LLM observabi
0	mit	sparowjake-troll/promptfoo	OURS	2026-07-25	Test your prompts, agents, and RAGs. Red teaming/pentesting/vulnerability scanning for AI. Compare performance
0	agpl-3.0 copyleft	sparowjake-troll/T3MP3ST	OURS	2026-07-05	autonomous red teaming platform; multi-agent offensive-security meta-harness

THE MATRIX — process × technique × tool × model × eval × autonomy

102 processes · 16 departments · 7 agent roles. BUILT 28 · WIRED 23 · TOOL 13 · PICK 2 · NONE 36
BUILT runs and is measured · WIRED runs, unmeasured · TOOL we own it, nothing wired · PICK identified, not obtained · NONE genuine gap. Autonomy: R read-only · D drafts, human releases · W writes to our own systems · X acts outside, needs sign-off.

Dept	Sub	Role	Process	Tier	Open-source tool	Our model	State (measured)	Aut	Artifact home	Gate
AI Infra	Eval	auditor	ai.gold_gate — block a collapsed model from shipping	—	evals/denial_gold.py (ours)	own model	BUILT exits non-zero; caught a 92% collapse	R	s3://getmax-dataroom/06_ai/evals/
AI Infra	Routing	manager	ai.route — cheapest model that passes the eval	—	LiteLLM (MIT) + our promptfoo matrix	own model grades	BUILT 7 apps x 4 providers	W	s3://getmax-dataroom/06_ai/evals/
AI Infra	Serving	—	ai.serve_brain — text+vision on our own L40S	—	vLLM (Apache) + Qwen3-Omni-30B-AWQ4	max-brain	BUILT 0.8s, ai.getmaxsolutions.com metered per call	W	—
AI Infra	Serving	—	ai.serve_ears — speech to text	—	faster-whisper + speaches	max-ears	BUILT round-trip proven	W	s3://getmax-dataroom/06_ai/transcripts/
AI Infra	Serving	—	ai.serve_voice — text to speech	—	Kokoro-FastAPI (Apache)	max-voice	BUILT 30KB mp3 proven	W	s3://getmax-dataroom/06_ai/audio/
AI Infra	Cost	analyst	ai.cost_per_run — what each call costs us	—	gmx-gateway usage.db	—	WIRED logging, nothing reads it	R	s3://getmax-dataroom/03_finance/ai_cost/
AI Infra	Serving	—	ai.serve_eyes — screenshot to click	T3	vLLM + UI-TARS-2B	max-eyes	WIRED — DOWN, nothing on :8001	W	—	restore uitars
AI Infra	Training	—	ai.train_slm — our own model from scored outputs	—	unsloth/axolotl (Apache) + getmax-slm	future SLM	NONE needs the corpus the loops produce	D	s3://getmax-dataroom/06_ai/models/	GPU window
Brand	Collateral	—	brand.pdf — one-pagers, proposals	—	HTML + headless Chromium print-to-PDF (no LaTeX needed)	max-brain	WIRED Suresh_OnePager.docx exists; not templated	D	s3://getmax-dataroom/05_brand/pdf/
Brand	Deck	—	brand.ppt_build — a deck from a brief	—	python-pptx + our template	max-brain	WIRED us-expansion/_source/make_suresh_deck.py is the proof it works	D	s3://getmax-dataroom/05_brand/ppt/decks/
Brand	Logo	—	brand.logo — marks and lockups	—	HTML+headless Chromium beats diffusion for type; Inkscape (GPL) for vector	max-brain writes the spec	WIRED logo.getmaxglobal.com exists (basic auth)	D	s3://getmax-dataroom/05_brand/logo/
Brand	Deck	—	brand.ppt_template — the reusable master	—	python-pptx (BSD) — templates as CODE, not files to lose	max-brain	TOOL python-pptx used before (Suresh_Deck.pptx)	D	s3://getmax-dataroom/05_brand/ppt/templates/
Brand	Assets	auditor	brand.asset_register — one home, no 4th brand folder	—	FOLDER_MAP.md + the data room	—	NONE — 4 brand folders + 3 video folders already exist	R	s3://getmax-dataroom/05_brand/
Brand	Collateral	—	brand.flyer — pamphlets, flyers, posters	—	HTML/CSS + Chromium; Inkscape (GPL) for print vector	max-brain	NONE	D	s3://getmax-dataroom/05_brand/collateral/
Client Ops	Onboard	manager	co.intake — who they are, what they run, who touches it	T0->T1	our bd.* chain + Docling (MIT) for their docs	max-brain	WIRED bd.understand_business is step 1 of this	D	s3://getmax-dataroom/13_clientops/<client>/intake/
Client Ops	Talk	—	co.speak — talk to THEIR customers, in voice	—	TEN Agent (ON-BOX) + LiveKit (ON-BOX) + our ears/voice	max-voice + max-ears	WIRED both running on gpu1, no URL, unused	X	s3://getmax-dataroom/13_clientops/<client>/calls/	client sign-off
Client Ops	Memory	—	co.remember — per-client long-term memory an agent can use	—	Letta (CLONED) + Qdrant (Apache)	max-brain	TOOL letta cloned, qdrant not obtained	W	s3://getmax-dataroom/13_clientops/<client>/memory/
Client Ops	Connect	—	co.connect — get into THEIR tools (portal, EHR, sheets, mail)	T1->T3	Activepieces (MIT) for APIs; UI-TARS for portals with no API	max-eyes	NONE — the single highest-value gap	X	s3://getmax-dataroom/13_clientops/<client>/connections/	client credentials + written consent
Client Ops	Harvest	research	co.harvest — pull knowledge off every device/tool we touch	T0-T3	Docling + PaddleOCR + our ears for calls	max-brain + max-ears	NONE	R	s3://getmax-dataroom/13_clientops/<client>/knowledge/	consent
Client Ops	Improve	auditor	co.improve — measure against the KPI, refine the SOP	—	promptfoo rows + GrowthBook (FORKED) for A/B	own model grades	NONE	W	s3://getmax-dataroom/13_clientops/<client>/evals/
Client Ops	KPI	analyst	co.set_kpi — the numbers that say the SOP is working	—	our own dbs + Metabase/Evidence	max-brain	NONE	D	s3://getmax-dataroom/13_clientops/<client>/kpi/
Client Ops	Observe	research	co.observe — watch how the work is actually done	T3	neko (Apache) to watch+take over; mitmproxy (MIT) to see their API calls	max-eyes + max-brain	NONE	R	s3://getmax-dataroom/13_clientops/<client>/observations/	consent
Client Ops	Pipeline	—	co.pipeline — a per-client data pipeline built at the instance	—	Activepieces + S3 + our own models	max-brain	NONE — generated per client, never hardcoded	W	s3://getmax-dataroom/13_clientops/<client>/pipeline/
Client Ops	Run	—	co.run — the agent executes the SOP, not a human	T1-T3	Activepieces + UI-TARS + our brain	max-brain + max-eyes	NONE — the endpoint of the whole plan	X	s3://getmax-dataroom/13_clientops/<client>/runs/	per-client sign-off
Client Ops	SOP	writer	co.write_sop — turn what we observed into the written SOP	—	open-notebook (CLONED) for retrieval over the harvest	max-brain	NONE — this is where the value compounds	D	s3://getmax-dataroom/13_clientops/<client>/sop/
Cust Support	Intake	—	cs.inbox — one queue for every channel	—	chatwoot LIVE support.getmaxsolutions.com (MIT) — email+chat+WhatsApp in one	max-brain	WIRED chatwoot 302->login, 99 tables; no channel connected yet	W	s3://getmax-dataroom/08_support/tickets/	SMTP key for the email channel; WhatsApp can use our own gowa/wapp gateway
Cust Support	Voice	—	cs.call_notes — support calls to notes	—	our own ears	max-ears + max-brain	WIRED meeting pipeline already does exactly this	W	s3://getmax-dataroom/08_support/calls/
Cust Support	Reply	writer	cs.draft_reply — answer from our own KB	—	open-notebook (CLONED oss/ai-rag) for retrieval	max-brain	TOOL cloned, not wired	D	s3://getmax-dataroom/08_support/replies/
Cust Support	QA	auditor	cs.qa — was the answer right and on-voice	—	promptfoo rows	own model grades	NONE	R	s3://getmax-dataroom/06_ai/evals/
Cust Support	Triage	analyst	cs.triage — category, urgency, owner	—	no tool needed	max-brain	NONE	D	s3://getmax-dataroom/08_support/tickets/
Deal Desk	Brief	writer	trade.digest — desk to Sriram's phone	—	getmax-desk digest + wapp	max-brain	BUILT delivered msg 3EB0B18453144FE688F2C3 2026-08-06	W	~/getmax-desk/desk.db
Deal Desk	Grading	analyst	trade.grade — spec-driven ISRI grade + Rs/kg Cu	—	our rules over seller spec attributes	max-brain	BUILT 20 of 46 priced listings credible, 8 flagged bad data	R	~/getmax-desk/desk.db
Deal Desk	Pricing	analyst	trade.benchmark — world price in Rs/kg	T0	COMEX HG=F + frankfurter FX, both keyless	—	BUILT Rs 1,408/kg cathode 2026-08-06 (COMEX = upper bound, US tariff)	R	~/getmax-desk/desk.db
Deal Desk	Sourcing	research	trade.source — find the supply side	T1	IndiaMART __INITIAL_STATE__ parser (ours)	—	BUILT 72 Chennai copper suppliers, 28 non-copper rejected 2026-08-06	R	~/getmax-desk/desk.db
Deal Desk	Outreach	manager	trade.rfq — RFQ out, quote back	—	wapp gateway (live) / voice (none) / sms (none)	max-brain	NONE — 0 of 72 suppliers WhatsApp-reachable, all IndiaMART PNS proxies	D	~/getmax-desk/desk.db	outbound PSTN carrier
Finance	Watch	auditor	fin.domain_watch — renewals before they lapse	—	scripts/domain_watch.py (ours)	—	BUILT cron 3:47 on gpu1, Brevo send proven (201)	W	s3://getmax-dataroom/03_finance/renewals/
Finance	Cost	analyst	fin.cloud_cost — AWS burn vs credits	—	aws ce + our own scripts	max-brain	WIRED credits tracked by hand; console is the only source	R	s3://getmax-dataroom/03_finance/cloud/
Finance	Reporting	analyst	fin.pack — monthly numbers, one pack	—	openpyxl + Chromium PDF	max-brain	WIRED CFO system exists in Finance folder	D	s3://getmax-dataroom/03_finance/reports/
Finance	Billing	—	fin.invoice — raise and track	—	PICK invoiceninja (FORKED) / midday (AGPL)	max-brain	TOOL forks exist, nothing wired	D	s3://getmax-dataroom/03_finance/invoices/
Finance	Audit	auditor	fin.recon — bank vs books	—	PICK midday (AGPL) does reconciliation	max-brain	NONE	D	s3://getmax-dataroom/03_finance/recon/
Finance	Collections	manager	fin.chase — follow up unpaid	—	reuse sales.sequence	max-brain drafts	NONE	D	s3://getmax-dataroom/03_finance/collections/
Founders	Meetings	—	fnd.meeting_pipeline — every recording to notes	—	faster-whisper + our brain, hourly cron	max-ears + max-brain	BUILT 94-min Suresh meeting summarised 2026-08-05 01:49	W	s3://getmax-dataroom/01_founders/meetings/
Founders	Register	manager	fnd.open_items — what is actually open	—	OPEN_ITEMS.md + product_asks	max-brain	BUILT	W	s3://getmax-dataroom/01_founders/register/
Founders	Board	analyst	fnd.metrics — the numbers that matter monthly	—	our own dbs + openpyxl	max-brain	NONE	D	s3://getmax-dataroom/01_founders/metrics/
Founders	Brief	analyst	fnd.daily_brief — one page, every morning	—	our own brain over the registers	max-brain	NONE — this is the 'keep me updated' row	D	s3://getmax-dataroom/01_founders/briefs/
Founders	Data room	—	fnd.dataroom — investor/DD-ready structure	—	S3 + a read view on getmaxsolutions.com	—	NONE — the tree is defined below, not created	W	s3://getmax-dataroom/	approve tree
IT	Mail	auditor	it.deliverability — can our mail actually land	—	DNS/SPF/DKIM checks + blocklist lookups	—	BUILT Barracuda LISTS our IP; relay is the fix, not delisting	R	s3://getmax-dataroom/07_tooling/mail/
IT	Uptime	auditor	it.verify_live — is each service really up	—	verify_live.py (ours) + netdata (CLONED)	—	BUILT but it MISSED 4 dead endpoints — coverage is the bug	W	s3://getmax-dataroom/07_tooling/uptime/
IT	Access	manager	it.access — who has what, joiner/leaver	—	onecli credential gateway + vault (RUNNING gpu1)	—	WIRED running, unused	W	s3://getmax-dataroom/07_tooling/access/
IT	Endpoint	auditor	it.endpoint — machines, patching	—	meshcentral (CLONED oss/monitoring)	—	TOOL	W	s3://getmax-dataroom/07_tooling/endpoints/
Legal	Contracts	writer	legal.draft — first draft from a template	—	python-docx + our templates	max-brain drafts	WIRED Suresh_Agreement_DRAFT.docx exists	D	s3://getmax-dataroom/12_legal/contracts/
Legal	Sign	—	legal.esign — send for signature	—	docuseal (CLONED) / documenso (AGPL)	—	TOOL cloned	X	s3://getmax-dataroom/12_legal/signed/	Sriram signs, never an agent
Legal	Compliance	auditor	legal.hipaa — no PHI where it should not be	—	our own scanners	max-brain	NONE — the rule exists, no automated check	R	s3://getmax-dataroom/12_legal/compliance/
Marketing	Content	writer	content.article — long-form on our own site	—	ContentPilot (OURS, live cp.getmaxglobal.com)	max-brain	BUILT live; anti-AI-voice row passes	D	s3://getmax-dataroom/04_marketing/content/articles/
Marketing	Content	copy writer	content.copy — landing/ad/email copy	—	our anti-AI-voice ban list as a hard assert	max-brain	BUILT it is a passing row in ai-lab	D	s3://getmax-dataroom/04_marketing/content/copy/
Marketing	Content	auditor	content.voice_audit — does it sound like a human	—	promptfoo row: banned-words + rubric	own model grades	BUILT passes on all 4 providers	R	s3://getmax-dataroom/04_marketing/content/audits/
Marketing	Competitor	research	mkt.competitor_positioning — how they pitch	T0	our bd.* fetch chain	max-brain	WIRED bd.* does this, needs a rubric	R	s3://getmax-dataroom/04_marketing/competitors/
Marketing	Social	manager	social.queue — schedule and post	—	social-engine (OURS); PICK postiz/mixpost (AGPL)	max-brain drafts	WIRED 49 posts paused since 05-13	D	s3://getmax-dataroom/04_marketing/social/	social pages
Marketing	Video	—	video.render — variants from a brief	—	ffmpeg + Pillow + Playwright + Kokoro	max-voice + judge	WIRED — the OTHER window owns this lane	D	s3://getmax-dataroom/05_brand/video/	VRAM
Marketing	Content	ghost writer	content.ghost — in Sriram's voice, first person	—	voice-builder + post-writer skills (ours)	max-brain	TOOL skills exist, not wired to the brain	D	s3://getmax-dataroom/04_marketing/content/ghost/
Marketing	Digital/SEO	research	mkt.seo_research — keywords, SERP, competitors	T0	PICK searxng (AGPL) + crawlseo (MIT, FORKED)	max-brain	TOOL crawlseo forked, never run	R	s3://getmax-dataroom/04_marketing/seo/research/
Marketing	Digital/SEO	analyst	mkt.seo_audit — our own sites, ranked fixes	T1	lighthouse + crawlseo	max-brain	TOOL	R	s3://getmax-dataroom/04_marketing/seo/audits/
Marketing	Competitor	analyst	mkt.competitor_pricing — track pricing over time	T0+M	PICK changedetection.io (Apache)	max-brain	PICK	R	s3://getmax-dataroom/04_marketing/competitors/pricing/
Marketing	Digital/AEO	research	mkt.aeo_audit — are WE the answer an LLM gives	—	our own brain + a question set (no tool needed)	max-brain	NONE — AEO is answer-engine optimisation; nothing built	R	s3://getmax-dataroom/04_marketing/aeo/
Marketing	Digital/AEO	analyst	mkt.aeo_gap — which questions we lose and why	—	our own brain, scored like any other eval row	max-brain	NONE	R	s3://getmax-dataroom/04_marketing/aeo/
Marketing	Digital/SEO	manager	mkt.seo_plan — what to fix in what order	—	our own matrix	max-brain	NONE	D	s3://getmax-dataroom/04_marketing/seo/plans/
Marketing	Digital/SEO	auditor	mkt.seo_verify — did the fix move anything	T0	changedetection.io (Apache) on our own SERP positions	max-brain	NONE	R	s3://getmax-dataroom/04_marketing/seo/verify/
Marketing	Video	auditor	video.score — hook, brand fit, CTA	—	promptfoo creative rows	own model grades	NONE — planned as rows in the SAME config	R	s3://getmax-dataroom/06_ai/evals/
People/HR	Onboarding	manager	hr.onboard — accounts, access, checklist	—	M365 Graph + onecli vault (RUNNING on gpu1)	max-brain	WIRED onecli exists, no HR flow on it	W	s3://getmax-dataroom/09_people/onboarding/
People/HR	Payroll	—	hr.payroll — run and record	—	hrms.db (ours)	—	WIRED agent exists in the RCM agents folder	D	s3://getmax-dataroom/09_people/payroll/
People/HR	Time	analyst	hr.activity — where time actually goes	—	activitywatch + meshcentral (CLONED)	—	TOOL cloned, not wired	W	s3://getmax-dataroom/09_people/activity/	consent
People/HR	Audit	auditor	hr.policy_audit — are we compliant	—	our own KB	max-brain	NONE	R	s3://getmax-dataroom/09_people/policy/
People/HR	Hiring	research	hr.source — find candidates	T2	PICK browser-use	max-brain	NONE	R	s3://getmax-dataroom/09_people/hiring/	ToS
People/HR	Hiring	analyst	hr.screen_cv — CV against the role	—	no tool needed; PDF via marker/MinerU	max-brain vision	NONE	D	s3://getmax-dataroom/09_people/hiring/
Product	Asks	analyst	prod.extract_asks — what users actually asked for	—	our meeting pipeline	max-brain	BUILT product_asks in Supabase; 18 asks vs 2 by hand	W	s3://getmax-dataroom/11_product/asks/
Product	Release	manager	prod.ship — branch, preview, promote	—	gh + vercel CLI	—	BUILT but Build45 has NO auto-deploy — every ship is manual	X	s3://getmax-dataroom/11_product/releases/
Product	QA	auditor	prod.web_test — click every button	T1->T3	playwright (installed) + PICK midscene (MIT, STARRED)	max-brain + max-eyes	WIRED /web-test skill exists	R	s3://getmax-dataroom/11_product/qa/
R&D	Bench	analyst	rnd.bench — does it beat what we serve	—	our own gold sets + promptfoo	own model grades	BUILT the denial gold gate is exactly this	R	s3://getmax-dataroom/06_ai/evals/
R&D	Scan	research	rnd.model_watch — what shipped this week	—	HF hub API + our landscape script	max-brain	WIRED done by hand; Qwen3.8-27B ~Aug 10 is the next grab	R	s3://getmax-dataroom/10_rnd/models/
R&D	Spike	—	rnd.spike — try it in a sandbox, keep or kill	T4	PICK e2b / daytona (both permissive)	max-brain	NONE	W	s3://getmax-dataroom/10_rnd/spikes/
Sales	BD	research	bd.understand_business — URL to profile + ICP	T0->T1	requests+bs4, playwright (installed)	max-brain @solutions	BUILT 4/4 correct; evidence verified in code	R	s3://getmax-dataroom/02_sales/profiles/
Sales	BD	auditor	bd.icp_rubric — vendor vs provider gate	—	promptfoo (MIT)	own model grades	BUILT gate 0 after athenahealth 95->25	R	s3://getmax-dataroom/06_ai/evals/
Sales	Outreach	manager	sales.sequence — first touch	—	listmonk LIVE mail.getmaxsolutions.com (AGPL); Brevo transactional API in vault	max-brain drafts	WIRED listmonk 200; 141 sent 51% open; 5,047 ready leads unsent	D	s3://getmax-dataroom/02_sales/outreach/	SMTP key (Brevo dashboard-only) for listmonk; API path already works
Sales	BD	analyst	bd.find_similar — one good customer to a list	T0	google-maps-scraper (AGPL, CLONED)	max-brain	TOOL cloned, never run	R	s3://getmax-dataroom/02_sales/lists/
Sales	Close	—	sales.proposal — quote and proposal	—	python-pptx + Chromium PDF; docuseal (CLONED) to sign	max-brain drafts	TOOL	D	s3://getmax-dataroom/02_sales/proposals/
Sales	Prospect	research	sales.find_people — who to contact	T2	PICK browser-use / stagehand (MIT)	max-brain	PICK	R	s3://getmax-dataroom/02_sales/contacts/	LinkedIn ToS
Sales	Verify	auditor	sales.verify_contact — does the mailbox exist	T0	NONE found in 711; smtplib+dnspython is the floor	—	NONE — B3 empty	R	s3://getmax-dataroom/02_sales/contacts/
Tooling	Browser	—	tool.browse — the technique ladder T0..T4	T0-T4	INSTALLED on gpu1: trafilatura/readability/bs4/lxml (T0) · playwright+chromium 151 (T1) · browser-use 0.13.7 (T2) · curl_cffi (A6)	max-brain + max-eyes	BUILT T0/T1/T2 proven on gpu1 2026-08-05 — browser-use returned 'Example Domain' driven by OUR model; T3 DOWN (docker blocked)	R	s3://getmax-dataroom/07_tooling/browser/
Tooling	Inventory	auditor	tool.inventory — what OSS we actually own	—	gh_oss_inventory.py + browser_landscape.py (ours)	—	BUILT 711 classified, 149 verified	R	s3://getmax-dataroom/07_tooling/inventory/
Tooling	Licence	auditor	tool.licence_check — can we ship it to a client	—	our SPDX check in the landscape	—	BUILT 112/149 permissive, AGPL flagged	R	s3://getmax-dataroom/07_tooling/licences/
Tooling	Secrets	auditor	tool.vault — one place for every key	—	~/.env.migration + SSM SecureString	—	BUILT never put a key in an SSM send-command	R	—
Tooling	Crons	manager	tool.crons — what runs on a schedule, and did it	—	cron + flock on gpu1	—	WIRED 3 crons (meetings :17, domain_watch 3:47, asks) — no single view	W	s3://getmax-dataroom/07_tooling/crons/
WorkMax	Answer	—	wm.concierge_web — quote, qualify, never invent	T1	workmax/api/chat.js, 5 tools + groundReply()	max-brain (ai.getmaxsolutions.com)	BUILT live 08-10: ₹4,999 + GST-extra returned grounded, pricing card	W	~/workmax/api/chat.js
WorkMax	Measure	auditor	wm.funnel — enquiries → tours → seats, weekly	T0	scripts/db.py + walk_app.py	—	BUILT walk PASS 08-10 (was 18 route-500s); funnel numbers still 0/0/0	R	~/workmax/marketing/TRACKER.md
WorkMax	Qualify	analyst	wm.qualify — seats/shift/start-date/budget onto the row	T1	concierge tool output → wm_enquiries	max-brain	WIRED — agent collects it, nothing writes it back	W	wm_enquiries
WorkMax	Capture	manager	wm.listings — submit + keep 19 portals live	T2	browser-use on gpu1 + LISTINGS_PACK_ANTONY.md master NAP	max-brain	TOOL — browser agent proven 08-05, no WorkMax recipe written	X	~/workmax/marketing/LISTINGS_PACK_ANTONY.md	per-portal account + OTP
WorkMax	Answer	—	wm.speed_to_lead — reply to every enquiry in <60s	T1	wm_enquiries trigger + Brevo (key in vault)	max-brain	NONE — capture works, nothing replies. 0 enquiries ever, so 0 misses so far	W	wm_enquiries
WorkMax	Book	—	wm.tour_book — offer 2 slots, write it, remind on WhatsApp	T1	wm_bookings + wapp gateway	max-brain	NONE — 0 bookings ever through the app	W	wm_bookings
WorkMax	Capture	—	wm.map_presence — be findable in 'coworking near me'	T0	Google Business Profile (listing exists, unclaimed, no phone)	—	NONE — #1 faucet, still shut. Verified unclaimed by scrape 08-01	X	~/workmax/docs/17_LEAD_ENGINE_AND_APIS.md	GBP claim + video/postcard verify
WorkMax	Close	—	wm.collect — take the first rupee	T0	Razorpay (app/lib/razorpay.ts wired)	—	NONE — no RAZORPAY_* in Vercel env or ~/.env.migration. ₹0 collected ever	X	wm_invoices	Razorpay merchant account
WorkMax	Follow-up	writer	wm.nurture — every enquiry chased until yes or no	T1	wm_enquiries + wapp/Brevo, draft-first	max-brain	NONE — the leak that kills coworking pipelines	D	wm_enquiries
WorkMax	Outbound	writer	wm.night_shift — the moat: US-hours RCM/BPO teams	T1	Sriram's own RCM network + OSM Overpass 5km, NOT scraped lists	max-brain	NONE — the 101 scraped rows are retired: 57 no email, 35 no phone, 0 named humans	D	wm_leads	named contacts only Sriram has

Role coverage — the same roles repeat across departments

research	10	2	4
analyst	18	4	6
manager	13	3	3
auditor	21	10	9
writer	7	2	3
ghost writer	1	0	0
copy writer	1	1	0
—	31	6	11

By department

AI Infra	8	5	2	0	0	1
Brand	6	0	3	1	0	2
Client Ops	11	0	2	1	0	8
Cust Support	5	0	2	1	0	2
Deal Desk	5	4	0	0	0	1
Finance	6	1	2	1	0	2
Founders	5	2	0	0	0	3
IT	4	2	1	1	0	0
Legal	3	0	1	1	0	1
Marketing	15	3	3	3	1	5
People/HR	6	0	2	1	0	3
Product	3	2	1	0	0	0
R&D	3	1	1	0	0	1
Sales	7	2	1	2	1	1
Tooling	5	4	1	0	0	0
WorkMax	10	2	1	1	0	6

The org — who owns what, judged on what, checked how

62 roles · 14 departments · 55 sub-departments. An agent can hold **15** of these today, **33** with a human release, and **14** need a human.
The matrix is the machinery; this is the accountability. Seniority: L1 runs a written SOP · L2 runs it and writes it · L3 owns a sub-dept and sets the KPI · L4 owns a dept · L5 founder. Every role carries its KPI, its deliverable, and the QMS gate that makes the output trustworthy.

Dept	Sub	Role	L	KPI	Cadence	Deliverable	SOP / process	QMS gate	Agent?	Model
Founders	Strategy	Founder/CEO	L5	revenue booked; runway months	monthly	the decision, written down	—	none — this is the final gate	NO	human judgement; no model
Founders	Comms	Chief of Staff (agent)	L3	hours from meeting to written actions	per meeting	meeting brief + action list	meet.transcribe + meet.extract_asks	every action carries a verbatim quote + timestamp	YES	max-ears + max-brain
Founders	Register	Ops Auditor (agent)	L3	open items with a named owner and date	daily	OPEN_ITEMS.md + product_asks rows	os.open_items	a row without a command that was run is marked claimed	YES	max-brain
Founders	Brief	Analyst (agent)	L2	one page, in his hands before 8am	daily	the daily brief	fnd.daily_brief — NOT BUILT	every number links to the query that produced it	PART	max-brain
AI Infra	Serving	Platform Lead (agent)	L4	p50 latency; uptime; cost per 1k calls	continuous	a live endpoint	ai.serve_brain	prove from the laptop with real cert validation, no -k	YES	—
AI Infra	Routing	Router (agent)	L3	% of processes on their measured-best model	weekly	the routing table	ai.route	no cell trusted without --repeat 5	YES	own model grades
AI Infra	Eval	Quality Auditor (agent)	L3	gold-set accuracy AND collapse below 60%	per model change	the gate verdict	ai.gold_gate	exit non-zero blocks the ship; accuracy alone is not enough	YES	own model grades
AI Infra	Cost	Cost Analyst (agent)	L2	$ per process run; credit burn vs budget	weekly	the cost table	ai.cost_per_run — gateway logs it, nothing reads it	cross-check against the AWS console, never CE alone	PART	max-brain
AI Infra	Training	ML Engineer	L4	does our own model beat what we serve	per corpus	a fine-tuned checkpoint	ai.train_slm — NOT BUILT	must beat the always-guess floor before it replaces anything	NO	needs a GPU window + human sign-off
Tooling	Inventory	Tool Auditor (agent)	L2	% of tools with a verified licence	monthly	the inventory + landscape	tool.inventory	a fork you never cloned is not a capability	YES	—
Tooling	Licence	Compliance Auditor (agent)	L3	zero copyleft tools in anything client-facing	per adoption	the licence verdict	tool.licence_check	AGPL/SSPL self-host only — never shipped	YES	—
Tooling	Platform	Infra Engineer (agent)	L3	services LIVE vs PICK; disk headroom	weekly	a running service on a URL	stack.py --plan	DNS before the container, every time	PART	—
Client Ops	Onboard	Onboarding Manager (agent)	L3	days from signature to first automated run	per client	the client profile + plan	co.intake	every claim carries a verbatim quote from their own material	PART	max-brain
Client Ops	Connect	Integration Engineer (agent)	L3	their systems reachable without a human	per client	working connections	co.connect — NOT BUILT	written consent on file before ANY credential is used	NO	client credentials are a human handover
Client Ops	Observe	Process Researcher (agent)	L2	steps captured vs steps they actually do	per client	the observation log	co.observe — NOT BUILT	consent on file; no PHI in the log	NO	max-eyes
Client Ops	SOP	SOP Writer (agent)	L2	% of observed steps covered by the SOP	per client	the written SOP	co.write_sop — NOT BUILT	a human who does the job signs it off before it runs	PART	max-brain
Client Ops	KPI	Performance Analyst (agent)	L3	agent throughput vs the human baseline	weekly	the KPI dashboard	co.set_kpi — NOT BUILT	baseline measured BEFORE the agent runs, or there is no claim	PART	max-brain
Client Ops	Run	Operator (agent)	L1	tasks completed; exception rate	continuous	the completed work	co.run — NOT BUILT	exceptions go to a human queue, never silently retried	NO	max-brain + max-eyes
Client Ops	Improve	Improvement Auditor (agent)	L3	KPI delta per SOP revision	monthly	the revised SOP + the evidence	co.improve — NOT BUILT	A/B before adopting; no change on a single run	PART	own model grades
Sales	BD	BD Researcher (agent)	L1	qualified profiles per day	daily	the ICP-scored profile	bd.understand_business	every field carries a quote we can find in the source	YES	max-brain
Sales	BD	BD Analyst (agent)	L2	% of profiles that convert to a reply	weekly	the ranked list	bd.find_similar — TOOL only	vendor vs provider gate must pass	PART	max-brain
Sales	Outreach	SDR (agent)	L1	replies per 100 sent; bounce rate	daily	sent sequences	sales.sequence — NOT BUILT	nothing sends until the mailbox is verified AND the relay is clean	NO	our IP is blocklisted; relay required
Sales	Close	Account Exec	L4	closed value; cycle length	monthly	the signed agreement	sales.proposal	Sriram signs. Never an agent.	NO	human
Marketing	Digital/SEO	SEO Researcher (agent)	L1	keywords mapped to a page	weekly	the keyword + gap list	mkt.seo_research — TOOL only	positions verified from a second source	PART	max-brain
Marketing	Digital/SEO	SEO Analyst (agent)	L2	technical issues found and fixed	weekly	the ranked audit	mkt.seo_audit — TOOL only	re-crawl after the fix or it is not fixed	PART	max-brain
Marketing	Digital/SEO	SEO Manager (agent)	L3	organic sessions; positions moved	monthly	the plan + the report	mkt.seo_plan — NOT BUILT	a plan without a measured before-state is a guess	NO	max-brain
Marketing	Digital/AEO	AEO Researcher (agent)	L2	% of buying questions where WE are the cited answer	monthly	the AEO gap report	mkt.aeo_audit — NOT BUILT	the question set is fixed before measuring, not after	PART	max-brain
Marketing	Content	Writer (agent)	L2	published pieces; time on page	weekly	the article	content.article	anti-AI-voice ban list is a hard assert	YES	max-brain
Marketing	Content	Ghost Writer (agent)	L2	engagement vs his own baseline	weekly	the post in his voice	content.ghost — TOOL only	he reads it before it posts. Always.	PART	max-brain
Marketing	Content	Copy Writer (agent)	L1	CTR; conversion on the page	per asset	headline + body + CTA	content.copy	banned-word assert + a human read-aloud test	YES	max-brain
Marketing	Content	Content Auditor (agent)	L3	% of output passing the voice gate	per piece	pass/fail + the reason	content.voice_audit	if you would not say it in a meeting, it fails	YES	own model grades
Marketing	Social	Social Manager (agent)	L3	reach; replies; follower delta	daily	the scheduled queue	social.queue — 49 posts PAUSED since 05-13	nothing posts without a human release	PART	max-brain
Marketing	Video	Video Producer (agent)	L2	variants shipped; hook retention	weekly	the rendered variant	video.render — the OTHER window owns this	scored before publish, not after	PART	max-voice + judge
Brand	Identity	Brand Owner	L4	consistency across every asset	quarterly	the brand system	brand.asset_register — NOT BUILT	ONE home. 4 brand folders already exist — that IS the failure	NO	human taste
Brand	Deck	Deck Builder (agent)	L1	decks produced per brief	per request	the .pptx	brand.ppt_build	from the template, never freehand	PART	max-brain
Brand	Collateral	Designer (agent)	L2	assets produced; rework rate	per request	PDF / flyer / one-pager	brand.pdf, brand.flyer	print-safe sizes; typography via HTML, never diffusion	PART	max-brain
Cust Support	Intake	Support Agent (agent)	L1	first-response time	continuous	the ticket, triaged	cs.inbox — NOT BUILT	no ticket closed without a resolution note	NO	needs Chatwoot deployed
Cust Support	Triage	Support Analyst (agent)	L2	% routed correctly first time	daily	category + owner + urgency	cs.triage — NOT BUILT	misroute rate reviewed weekly	PART	max-brain
Cust Support	Reply	Support Writer (agent)	L2	resolution rate without escalation	continuous	the reply	cs.draft_reply — TOOL only	answer must cite the KB passage it came from	PART	max-brain
Cust Support	QA	Support Auditor (agent)	L3	% of replies correct AND on-voice	weekly	the QA scorecard	cs.qa — NOT BUILT	sampled, not self-reported	PART	own model grades
People/HR	Hiring	Recruiter (agent)	L1	qualified candidates per role	per role	the shortlist	hr.source — NOT BUILT	no scraping that breaches a platform's ToS	NO	ToS gate
People/HR	Hiring	Hiring Analyst (agent)	L2	screen-to-interview conversion	per role	the scored CV set	hr.screen_cv — NOT BUILT	criteria written BEFORE seeing CVs; no protected attributes	PART	max-brain vision
People/HR	Onboarding	People Manager	L3	days to fully productive	per joiner	accounts + access + checklist done	hr.onboard — WIRED, unused	access reviewed at 30 days	PART	max-brain
People/HR	Time	People Analyst (agent)	L2	utilisation; overtime flags	weekly	the activity report	hr.activity — TOOL only	CONSENT first. Surveillance without it is not on the table.	NO	consent gate
People/HR	Policy	HR Auditor (agent)	L3	policies current and acknowledged	quarterly	the compliance statement	hr.policy_audit — NOT BUILT	a policy nobody signed is not a policy	PART	max-brain
Finance	Billing	Billing Clerk (agent)	L1	invoices raised on time	monthly	the invoice	fin.invoice — TOOL only	every invoice ties to a delivered outcome	PART	max-brain
Finance	Collections	Collections Manager (agent)	L3	DSO; overdue value	weekly	the chase sequence	fin.chase — NOT BUILT	tone reviewed by a human before send	PART	max-brain
Finance	Reporting	Finance Analyst (agent)	L2	pack out by day 5	monthly	the numbers pack	fin.pack — WIRED	every figure traceable to a source row	PART	max-brain
Finance	Cost	Cloud Cost Analyst (agent)	L2	burn vs credits; $/process	weekly	the cloud cost view	fin.cloud_cost — WIRED	console is the source of truth, not Cost Explorer	PART	max-brain
Finance	Watch	Renewal Auditor (agent)	L2	zero services lost to an expiry	daily	the alert	fin.domain_watch — BUILT, cron 3:47	fires on the DATE; auto-renew is invisible to every API	YES	—
Finance	Controls	Financial Controller	L4	clean reconciliation; no surprises	monthly	signed accounts	fin.recon — NOT BUILT	a human signs. Always.	NO	human
IT	Access	IT Manager (agent)	L3	zero orphan accounts	monthly	the access register	it.access — WIRED, unused	joiner/leaver same day	PART	—
IT	Uptime	SRE (agent)	L2	% of endpoints probed AND up	continuous	the uptime report	it.verify_live — BUILT but MISSED 4 dead endpoints	an endpoint outside the probe list is one nobody knows is down	PART	—
IT	Mail	Deliverability Auditor (agent)	L3	inbox placement; blocklist status	weekly	the deliverability verdict	it.deliverability — BUILT	residential dynamic IP means a relay, not a delisting request	YES	—
R&D	Scan	Research Analyst (agent)	L2	relevant releases caught within a week	weekly	the model watch list	rnd.model_watch — WIRED by hand	claims checked against the model card, not a tweet	PART	max-brain
R&D	Bench	Benchmark Engineer (agent)	L3	does it beat what we serve, on OUR set	per candidate	the bench result	rnd.bench — BUILT	our own gold set only; published numbers are not comparable	YES	own model grades
Product	QA	QA Auditor (agent)	L2	broken paths found before a user does	per release	the QA report	prod.web_test — WIRED	screenshots attached or it did not happen	PART	max-brain + max-eyes
Product	Asks	Product Analyst (agent)	L2	asks captured vs asks in the recording	per meeting	product_asks rows	prod.extract_asks — BUILT	verbatim quote + timestamp per ask	YES	max-brain
Product	Release	Release Manager	L3	ships per week; rollbacks	per ship	the promoted deployment	prod.ship — BUILT, manual	branch -> preview -> promote. Never overwrite a working version.	PART	—
Legal	Contracts	Legal Writer (agent)	L2	turnaround on a first draft	per request	the draft	legal.draft — WIRED	a human lawyer reviews anything binding	PART	max-brain
Legal	Compliance	Compliance Auditor (agent)	L3	zero PHI where it should not be	monthly	the compliance report	legal.hipaa — NOT BUILT	mock data in every test. No real patient data in code, ever.	PART	max-brain
Legal	Sign	Founder	L5	executed agreements	per deal	the signed contract	legal.esign — TOOL only	Sriram signs. Never an agent. No exceptions.	NO	human

Capacity — what it costs and whether it fits

• All 47 services on the list want 144 GB of disk and 100 GB of RAM if every one ran at once.
• gpu1 has 12 GB free. gpu2 has ~43 GB. Together 55 GB.
• So the whole stack does NOT fit on the GPU boxes. That is not a blocker — most of it does not belong there.
• Reclaim adds 97 GB to gpu1 (verified S3 mirrors only) → 109 GB free.
• kimi has 5.9 TB free and 247 GB RAM and no GPU. Every CPU-only service on this list belongs there, not on a GPU box. That single move solves the disk problem outright.

The fleet, read live

Box	Disk free	RAM	VRAM free	Admin path
gpu1 `i-055`	12 G of 310 (97%)	30 G	16,410 MiB	SSM ✅
gpu2 `i-025`	~43 G of 194	30 G	~5,000 MiB	SSM ✅
kimi	5.9 TB free	247 G	none (no GPU)	SSM ✅
engine	—	32 G	none	❌ no SSM agent, unreachable

Reclaim — verified, and one correction

Model	Local	Verdict
`gpt-oss-120b`	60 G	✅ delete — 35 objects / 182.3 GB in S3 — verified
`gpt-oss-20b`	13 G	✅ delete — 16 objects / 38.5 GB in S3 — verified
`ornith-35b`	24 G	✅ delete — BACKED UP 2026-08-05 on Sriram's instruction: 15 objects / 25,722,949,796 bytes at text/ornith-35b/, and the 22.1 GB Ornith-1.0-35B-UD-Q4_K_M.gguf matches local byte-for-byte
`qwen3-omni-awq4`	26 G	⛔ KEEP — mounted by vllm-omni, in use
`ui-tars-2b`	9.2 G	⛔ KEEP — T3 vision needs it
`hub`	30 G	⛔ KEEP — HF cache — re-downloadable, but that costs egress + time

Per-service sizing

Service	Disk GB	RAM GB	VRAM MiB	Basis	Note
vLLM max-brain (Qwen3-Omni-30B-AWQ4)	26	6	29042	M	in use on gpu1
FLUX.1-schnell	24	4	12000	P	fp8 ~12 GB VRAM; the other lane wants this
UI-TARS-2B (max-eyes)	9.2	3	11517	M	2B model; VRAM is the 0.25 util setting
PostgreSQL (ONE shared instance)	8	2	—	P	5 services each wanting their own PG is the difference between fitting and not
ComfyUI	5	4	5022	M	already holding VRAM on gpu2
Huly	4	3	—	P	PG + S3 + elastic-ish index
Qdrant	4	2	—	P	disk grows with vectors — 4G is a starting point
Penpot	4	4	—	P	PG + Redis + 4 services
Chatwoot	3.5	2.5	—	P	needs PG + Redis; sidekiq worker is the RAM
Twenty CRM	3	3	—	P	PG + Redis + worker
neko (watch + take over)	3	3	—	P	WebRTC + a full browser
Firecrawl	3	3	—	P	AGPL; needs PG + Redis
PaddleOCR	3	3	1500	P	GPU optional; CPU works, slower
Rocket.Chat	3	3	—	P	Mongo
TEN Agent	2.5	2	—	E	already on gpu1
Windmill	2.5	2	—	P	workers scale with concurrency
Steel Browser	2.5	4	—	P	RAM is per concurrent browser — 4G is ~4 tabs
Browserless	2.5	4	—	P	same shape; SSPL licence
Docling	2.5	3	—	P	model weights for layout inside the image
Langfuse	2.5	2	—	P	PG + clickhouse in v3 — check before deploying
Selenium Grid (hub + 2 nodes)	2	6	—	P	3G RAM per node is realistic
Crawl4AI	2	2.5	—	P	bundles a browser
Maxun	2	2.5	—	E	PG + browser
open-notebook	2	2	—	E	PG + embeddings
GrowthBook	2	2	—	P	wants Mongo AND PG
faster-whisper medium (max-ears)	1.5	3	4296	M	int8, 39-49x realtime
Activepieces	1.5	1.5	—	P	MIT; prefer over n8n
Metabase	1.5	2	—	P	JVM — 2G is the floor, not the target
Documenso / DocuSeal	1.5	1	—	P	one of the two, not both
Letta	1.5	2	—	P	PG
onecli + postgres	1.2	1	—	E	already on gpu1
n8n	1.2	1	—	P	NOT OSI — Sustainable-Use
Open WebUI	1.1	1.5	—	M	/opt/owui-data measured
OpenObserve	1	1.5	—	P	state on S3 — the cheap one
MeshCentral	1	1	—	P
Evidence	0.5	0.5	—	E	static build output, no server state
changedetection.io	0.5	0.7	—	P	+ a browser sidecar if JS sites
Netdata	0.5	0.5	—	P	per-second metrics, RAM ring buffer
Kokoro TTS	0.4	2	—	M	CPU only
LiveKit	0.3	1	—	E	RAM-heavy only under call load
Redis (shared)	0.3	0.5	—	P	chatwoot/langfuse/growthbook all want one
Listmonk	0.3	0.3	—	P	tiny; PG holds the lists
SearXNG	0.3	0.5	—	P	stateless, RAM only
ActivityWatch	0.3	0.3	—	P	runs on the endpoint, not the server
mitmproxy	0.2	0.4	—	P	internal only, never public
Caddy	0.05	0.2	—	M
gmx-gateway	0.01	0.3	—	M	FastAPI + SQLite

The unified stack — every service we deploy, one list

54 deployable services. LIVE 26 · ON-BOX (running, no URL) 6 · CLONED (on disk, not running) 7 · PICK (chosen, not obtained) 15
Replaces six scattered menus (CLI_STACK, CREATIVE_STACK, OSS_ARSENAL, CLAUDE_CODE_ARSENAL, oss/INDEX, powerups/SHORTLIST) — there was no single answer to “what is deployed and what is not”. Every target is a *.getmaxsolutions.com subdomain. A licence marked ⚠ is copyleft or source-available: fine to self-host, not fine to ship to a client.

Service	What it does	UI	Login	Licence	Store	Port	Departments	Status	Target URL	Blocker
A6: curl_cffi	Real browser TLS/JA3 impersonation	no	none	MIT	—	—	Tooling	LIVE	/opt/browser (venv)
Browserless	Browser-as-a-service: pooling, queues, session limits	YES debug UI	token	SSPL ⚠	RAM	3000	Tooling · Client Ops	LIVE	takeover.getmaxsolutions.com	verified 2026-08-06 LIVE behind basic auth (user sriram; pw in vault NEKO_USER_PASSWORD + SSM /getmax/neko/user-pw). 401 unauth, 200 authed, returned a real 19,343-byte PNG. Debugger UI = watch-and-drive; POST /screenshot//content//pdf//function = the agent path
Caddy	Reverse proxy + automatic Let's Encrypt	no	—	Apache	EBS	80/443	IT	LIVE	*	point DNS BEFORE starting a container
Chatwoot	One inbox: email + chat + WhatsApp + web widget. THE customer front door	YES	user accounts + teams	MIT	PG	3000	Support · Sales · Client Ops	LIVE	support.getmaxsolutions.com	verified 2026-08-06 HTTP 302 -> login. DNS repointed to kimi 3.145.44.100. Still needs an SMTP key for the email channel; WhatsApp can use our own gowa
DocuSeal	E-signature, lighter	YES	user accounts	AGPL ⚠	PG	3000	Legal · Sales	LIVE	sign.getmaxsolutions.com	verified 2026-08-06 HTTP 302 · fleet-docuseal on kimi · either/or with Documenso
Langflow	Visual LLM flow builder — what is ACTUALLY answering flow.	YES	user accounts	MIT	PG	7860	AI Infra · Client Ops	LIVE	flow.getmaxsolutions.com	verified 2026-08-09 HTTP 200 Server: uvicorn. This file previously claimed flow. was an unclaimed PICK for n8n/Activepieces. It is neither — Langflow holds the name
Langfuse	LLM tracing + eval UI — see every agent call	YES	user accounts	MIT core	PG	3000	AI Infra · all	LIVE	traces.getmaxsolutions.com	verified 2026-08-06 HTTP 200 · fleet-langfuse on kimi · URL is traces. not trace.
Letta (MemGPT)	Long-term agent memory — agents that remember a customer	YES	token	Apache	PG	8283	Client Ops · Support	LIVE	letta.getmaxsolutions.com	verified 2026-08-06 HTTP 307 · fleet-letta on kimi · URL is letta. not mem.
Listmonk	Self-hosted email campaigns + lists	YES	admin login	AGPL ⚠	PG	9000	Marketing · Sales	LIVE	mail.getmaxsolutions.com	verified 2026-08-06 HTTP 200 · comms-listmonk on kimi · SMTP still unset; Brevo API path works
MeshCentral	Remote machine management, patching	YES	user accounts	Apache	EBS	443	IT · People/HR	LIVE	mesh.getmaxsolutions.com	verified 2026-08-06 HTTP 302 · fleet-meshcentral on kimi · URL is mesh. not fleet.
Metabase	Point it at a DB, get dashboards non-engineers can use	YES	user accounts	AGPL ⚠	PG	3000	Finance · Founders · all	LIVE	bi.getmaxsolutions.com	verified 2026-08-09 DNS repointed gpu1 -> kimi 3.145.44.100; HTTP 200 with metabase.DEVICE cookie = the real Metabase, not a status-code guess. Name collision resolved
Open WebUI	Chat UI over our own models	YES	user accounts	MIT	PG	3000	all — the front door for staff	LIVE	chat.getmaxsolutions.com
Qdrant	Vector store the RAG layer needs	YES dashboard	API key	Apache	EBS	6333	AI Infra · Client Ops	LIVE	vec.getmaxsolutions.com	verified 2026-08-06 HTTP 401 = live behind basic auth · gmxstack-qdrant on gpu1:8183
SearXNG	Self-hosted meta-search — replaces a paid search API	YES	none	AGPL ⚠	RAM	8080	Marketing · Sales · R&D	LIVE	search.getmaxsolutions.com	verified 2026-08-09 DNS repointed gpu1 -> kimi 3.145.44.100; HTTP 200 Server: granian = the real SearXNG. Was 502 against a dead gpu1 listener
T0 stack: trafilatura+readability+bs4+lxml+markdownify	Fetch and convert, no browser, no model	no	none	MIT/Apache	EBS 265M	—	Tooling · Sales · Marketing · Client Ops	LIVE	/opt/browser (venv)	proven: trafilatura pulled 112 chars from a real page
T1: playwright + chromium 151	Real browser, deterministic selectors	no	none	Apache	EBS	—	Tooling · Product QA · Client Ops	LIVE	/opt/browser (venv)	proven: rendered GPU1 BROWSER OK; needs --no-sandbox as root
T2: browser-use 0.13.7	LLM reads the page and decides the next action. 107,870 stars	no	none	MIT	EBS 451M	—	Tooling · Sales · Client Ops	LIVE	/opt/browser311 (py3.11 venv)	PROVEN driving chromium with OUR model: returned 'Example Domain'
Twenty	Open CRM — the system of record for customers	YES	user accounts	AGPL ⚠	PG	3000	Sales · Client Ops	LIVE	crm.getmaxsolutions.com	verified 2026-08-06 HTTP 200. DNS repointed to kimi 3.145.44.100
changedetection.io	Watch a page/selector, diff it, alert	YES	password	Apache	EBS	5000	Marketing · R&D · Client Ops	LIVE	watch.getmaxsolutions.com	verified 2026-08-09 DNS repointed gpu1 -> kimi 3.145.44.100; HTTP 200 Server: Werkzeug/Python 3.11 = the real changedetection.io. Name collision resolved
gmx-gateway	Metered OpenAI-compatible gateway, per-call cost to SQLite	no	per-user key	ours	SQLITE	8080	AI Infra · Finance	LIVE	ai.getmaxsolutions.com
go-whatsapp-web-multidevice (gowa)	Our own WhatsApp Web API — the send/receive path for wapp	YES	basic auth	MIT	EBS	3000	Sales · Support · Client Ops	LIVE	wa.getmaxsolutions.com	verified 2026-08-09 HTTP 200. container fleet-gowa on kimi. Was serving all along, listed nowhere
google-maps-scraper	Places/leads scraped straight from Maps — the local-lead source	YES	none	MIT	EBS	8080	Sales · Marketing	LIVE	maps.getmaxsolutions.com	verified 2026-08-09 HTTP 200. container fleet-mapsscraper on kimi. Listed nowhere until today
open-notebook	RAG over our own docs — the SOP/knowledge brain	YES	user accounts	MIT	PG	8502	Client Ops · Support · People/HR	LIVE	kb.getmaxsolutions.com	verified 2026-08-06 HTTP 307 · agents-opennotebook on kimi
promptfoo	Prompt/agent regression tests	CLI + report	none	MIT	SQLITE	15500	AI Infra · all	LIVE	—	installed and running our matrix
python3.11.15	browser-use needs >=3.11; box python is 3.10.12	no	—	PSF	—	—	Tooling	LIVE	deadsnakes PPA, 3.10 left intact
vLLM (max-brain)	Serves our own text+vision model	no	API key	Apache	EBS+RAM	8002	AI Infra · all	LIVE	ai.getmaxsolutions.com
ComfyUI	Node graph for image/video generation	YES	none	GPL ⚠	EBS	8188	Brand · Marketing	ON-BOX	—	the OTHER window owns this lane
Kokoro-FastAPI	Text to speech, 54 voices	no	API key	Apache	EBS	8880	AI Infra · Marketing · Support	ON-BOX	hub.getmaxglobal.com ⚠	verified 2026-08-06 RECONCILED: container kokoro-tts is UP on gpu2 (i-025) 127.0.0.1:8880 per docker ps, not on gpu1. ai. has NO audio route. Only path today is hub.getmaxglobal, EXPIRES 22 AUG
LiveKit (voicelab)	WebRTC media server for realtime voice	no	token	Apache	RAM	7880	Support · Sales	ON-BOX	voicelab.getmaxsolutions.com
TEN Agent	Realtime conversational voice agent framework	YES	token	Apache	RAM	49483	Support · Sales	ON-BOX	ten.getmaxsolutions.com	auth gated, unused
onecli + postgres	Credential gateway with a built-in vault, scoped agent access	YES	user accounts	open	PG	10254	IT · People/HR	ON-BOX	—	no vhost yet
speaches / faster-whisper	Speech to text on the L40S	no	API key	MIT	EBS	8010	AI Infra · Support · Founders	ON-BOX	hub.getmaxglobal.com ⚠	verified 2026-08-06 RECONCILED: container murmur-ears is UP on gpu2 (i-025) 127.0.0.1:8010 per docker ps. It is NOT on gpu1 — the other window checked gpu1, found 000, and marked it PICK. Neither LIVE-at-ai nor PICK: it runs, but ai. has NO audio route (gateway patch staged at /tmp/app.py on gpu1). Only path today is hub.getmaxglobal, EXPIRES 22 AUG
ActivityWatch + aw-app	Where time actually goes	YES	local	MPL	SQLITE	5600	People/HR	CLONED	—	consent required before any rollout
GrowthBook	Feature flags + A/B — decides when an agent respawns	YES	user accounts	MIT	PG+Mongo	3100	C-Suite · Product	CLONED	ab.getmaxsolutions.com
Huly	Project/issue tracking, Jira replacement — Jira is dead on our tenant	YES	user accounts	EPL	PG+S3	8087	Product · all	CLONED	pm.getmaxsolutions.com
Netdata	Live per-second box metrics	YES	none	GPL ⚠	RAM	19999	IT	CLONED	metrics.getmaxsolutions.com
OpenObserve	Logs + traces + metrics in one, cheap on S3	YES	user accounts	AGPL ⚠	S3	5080	IT · AI Infra	CLONED	logs.getmaxsolutions.com
Rocket.Chat	Self-hosted team chat	YES	user accounts	MIT	Mongo	3000	Comms	CLONED	talk.getmaxsolutions.com
UI-TARS-2B (max-eyes)	Screenshot to click x,y — drives any portal	no	API key	Apache	EBS	8001	Tooling · Client Ops · Product	CLONED	—	verified 2026-08-06 STOPPED on purpose — browserless+playwright drives portals better and free. Freed 6.4 GB VRAM on gpu1 (37.1->30.7 used). Restart: docker start uitars-t3
Activepieces	MIT-licensed workflow runner — the safe alternative to n8n	YES	user accounts	MIT	PG	8080	Client Ops · all operations	PICK	flow.getmaxsolutions.com	either/or with n8n — Activepieces is MIT, prefer it
Crawl4AI	Site to LLM-ready markdown, as a service	API only	token	Apache	S3	11235	Tooling · Sales · Marketing	PICK	crawl.getmaxsolutions.com
Docling	Any document to structured markdown	API	none	MIT	S3	5001	Client Ops · Finance · Legal	PICK	doc.getmaxsolutions.com
Documenso	E-signature, self-hosted	YES	user accounts	AGPL ⚠	PG	3000	Legal · Sales	PICK	sign.getmaxsolutions.com
Evidence	BI as code — dashboards from SQL + markdown, versioned	no build UI	none	MIT	NONE	3000	Finance · Founders	PICK	bi.getmaxsolutions.com	either/or with Metabase
Firecrawl	Whole-site crawl to markdown at scale	YES	API key	AGPL ⚠	PG+S3	3002	Tooling · Marketing	PICK	crawl.getmaxsolutions.com	AGPL: self-host only, never ship
Maxun	No-code scraper that emits a reusable robot	YES	user accounts	AGPL ⚠	PG	5173	Tooling · Sales	PICK	robot.getmaxsolutions.com
PaddleOCR	OCR for screens and scans the DOM won't give us	API	none	Apache	EBS	8868	Tooling · Finance	PICK	ocr.getmaxsolutions.com
Penpot	Open Figma — design files the org can share	YES	user accounts	MPL	PG+S3	9001	Brand · Marketing	PICK	design.getmaxsolutions.com
Selenium Grid	Many parallel browsers, the scale-out standard	YES console	none	Apache	RAM	4444	Tooling · Product QA	PICK	grid.getmaxsolutions.com
Steel Browser	Open browser-as-an-API: sessions, proxies, captcha hooks	YES	token	Apache	RAM	3000	Tooling · Client Ops	PICK	browser.getmaxsolutions.com	either/or with Browserless — Steel is Apache, prefer it
Windmill	Turn a script into a UI + a cron. Developer-first ops	YES	user accounts	AGPL ⚠	PG	8000	Client Ops · IT	PICK	ops.getmaxsolutions.com
mitmproxy	See exactly what a customer's app calls — API discovery	YES web	none	MIT	EBS	8081	Tooling · Client Ops	PICK	mitm.getmaxsolutions.com	internal only, never public
n8n	Visual workflow runner, 400+ connectors — how we connect INTO a customer's tools	YES	user accounts	Sustainable-Use ⚠ NOT OSI	PG	5678	Client Ops · all operations	PICK	flow.getmaxsolutions.com	read the licence first
neko	A browser in a container you can WATCH and take over — human handoff	YES	room password	Apache	RAM	8080	Client Ops · Support	PICK	—	verified 2026-08-06 BLOCKED — docker pull/run of a new image over SSM is refused by the auto-mode classifier. Browserless took the takeover subdomain instead. Needs a Bash permission rule

How a row goes from PICK to LIVE

— how a row goes from PICK to LIVE on our own AWS, every time

  1. DNS FIRST, always:   python scripts/hostinger_dns.py set getmaxsolutions.com <name> A 3.147.149.16 --apply
     (Caddy cannot get a Let's Encrypt cert before the name resolves. Pointing DNS after
      starting the container is the single most common failure.)
  2. Storage by kind, not by habit:
       EBS   a database or a model that must survive a restart   (gpu1 has 14G free — TIGHT)
       S3    documents, transcripts, crawl output, backups        (cheap, already used)
       RAM   browsers and media servers — ephemeral by nature     (tmpfs, never persist)
       PG    one shared postgres beats one per app                (fewer containers, less disk)
  3. Container + 4 Caddy labels = automatic HTTPS. No manual cert work.
  4. Prove it from the LAPTOP with real cert validation (no -k) and grep the served bytes.
  5. Add the row to matrix.py with its artifact home, or the org cannot find its output.

⚠ gpu1 disk is at 96% (14G free). Anything with a big EBS footprint needs the 89 GB
  reclaim (gpt-oss-120b + ornith-35b, both backed up in S3) FIRST.
⚠ AGPL/SSPL rows are fine self-hosted for US. They are NOT fine shipped to a client.

Session claims — the anti-duplication lock

Two windows built brain.py twice in one day. A session now claims a process before touching it, and matrix.py --ready hides anything already claimed. The file is committed, so a session on another machine sees the same claims. A claim goes stale after 6 hours.

Process	Session	Status	When (UTC)	Note
ai.kimi_rightsize	session-30104017	working	2026-08-09T01:53	kimi i4i->r7i.2xlarge; NOT touching gpu1/gpu2/gpu3 - vllm-q35 and llama120 are another session's
tool.browse	sess-browser-0808	working	2026-08-07T23:08	browser visibility console + job runner slice 1
co.observe	sess-intake-0807	working	2026-08-07T16:26	OneDrive intake engine: call recordings -> transcript -> enrichment -> Excel
fin.chase	sess-comms	working	2026-08-05T01:15	comms/inbox lane - agent-native email OS
sales.sequence	sess-comms	working	2026-08-05T01:15	comms/inbox lane - agent-native email OS
cs.inbox	sess-comms	working	2026-08-05T01:15	comms/inbox lane - agent-native email OS
fnd.daily_brief	sess-comms	working	2026-08-05T01:15	comms/inbox lane - agent-native email OS

The gold-set gate — where accuracy alone lies

276 real denial lines with known answers. Collapse is the largest share any single label takes of our predictions — a model that answers one class for everything can post a respectable accuracy while being useless. Always-guess is what a hardcoded string scores. Anything that fails to beat it is worse than no model.

Provider	n	Accuracy	Always-guess	Collapse	Collapses to	Verdict
solutions/max-brain	8	37.5%	50.0%	88%	true_denial	FAIL
hub/max-brain-025	40	7.5%	40.0%	92%	contractual_adjustment	FAIL
hub/max-brain-055	40	10.0%	40.0%	85%	contractual_adjustment	FAIL
solutions/max-brain	40	35.0%	40.0%	92%	true_denial	FAIL
solutions/max-brain	40	35.0%	40.0%	92%	true_denial	FAIL

Read the top row carefully. The best accuracy in this table belongs to a provider that answers the SAME label for ~92% of inputs — it scores well only because that label happens to be the most common gold answer. It is a constant guesser. That is the entire reason this page reports collapse next to accuracy, and why no process here asks the model to judge without an evidence check.

The full browser-automation landscape — by technique

75 tools verified live against the GitHub API, not just the ones in our stars. Ours? distinguishes starred from forked from actually cloned — a fork you never cloned is not a capability. Tiers are cost tiers: T0 costs ~0, T3 costs a model call per step, so they are not substitutes.

T0 — Fetch + convert. No browser, no model. Cheapest thing that can work. 18 tools 4 ours

Repo	Stars	Licence	Pushed	Ours?	What ONLY this one gives you
unclecode/crawl4ai	76,057	Apache-2.0	2026-07-30	⬜ not ours	Markdown tuned for LLM input; chunking + extraction strategies built in
firecrawl/firecrawl	160,984	AGPL-3.0	2026-08-04	⬜ not ours	Whole-site crawl to markdown as a service you can self-host
D4Vinci/Scrapling	72,551	BSD-3-Clause	2026-07-30	⭐ starred	Adaptive selectors that survive a site redesign
adbar/trafilatura	6,403	Apache-2.0	2026-08-04	⬜ not ours	Best-in-class boilerplate removal; academic-grade text extraction
microsoft/markitdown	171,454	MIT	2026-07-29	⬜ not ours	Any file (pdf/docx/xlsx/pptx) to markdown, one API
scrapy/scrapy	63,638	BSD-3-Clause	2026-08-04	⬜ not ours	The mature crawl framework: scheduling, retries, pipelines, throttling
apify/crawlee	25,187	Apache-2.0	2026-08-04	⬜ not ours	Crawl framework that switches HTTP<->browser per request automatically
gocolly/colly	25,403	Apache-2.0	2026-06-18	⬜ not ours	Single-binary Go crawler; fastest per-page, no runtime
projectdiscovery/katana	17,269	MIT	2026-08-03	⬜ not ours	Crawler built for endpoint discovery, not content
buriy/python-readability	2,894	Apache-2.0	2026-01-26	⬜ not ours	The Readability port — small, predictable article extraction
codelucas/newspaper	15,135	MIT	2026-07-21	⬜ not ours	Article + author + publish-date extraction for news
Y2Z/monolith	15,379	CC0-1.0	2026-05-25	⬜ not ours	Freeze a page into ONE self-contained html file
gildas-lormeau/SingleFile	22,081	AGPL-3.0	2026-07-31	⬜ not ours	Same, as a browser extension — captures post-JS state
ArchiveBox/ArchiveBox	28,070	MIT	2026-08-03	⬜ not ours	Permanent self-hosted archive with many output formats
datalab-to/marker	38,342	Apache-2.0	2026-07-20	🍴 fork	PDF to markdown with layout + tables preserved
opendatalab/MinerU	76,772	NOASSERTION	2026-07-30	⭐ starred	PDF/scan to structured data incl. formulas
any4ai/AnyCrawl	3,383	MIT	2026-07-31	⭐ starred	Node crawler emitting LLM-ready output
dgtlmoon/changedetection.io	32,614	Apache-2.0	2026-08-04	⬜ not ours	Scheduled DIFF of a page — the competitor-pricing primitive

T1 — Real browser, deterministic selectors. Still no model. Where bot walls start. 14 tools 1 ours

Repo	Stars	Licence	Pushed	Ours?	What ONLY this one gives you
microsoft/playwright	93,978	Apache-2.0	2026-08-04	⬜ not ours	The reference driver: auto-wait, tracing, 3 engines, best debugging
puppeteer/puppeteer	95,409	Apache-2.0	2026-08-04	⬜ not ours	Closest to raw CDP; the Chrome-team lineage
SeleniumHQ/selenium	34,348	Apache-2.0	2026-08-04	⬜ not ours	The W3C standard + Grid; the only one with real enterprise/legacy reach
go-rod/rod	7,053	MIT	2026-07-15	⬜ not ours	Go, CDP-native, no Node runtime
chromedp/chromedp	13,229	MIT	2026-07-14	⬜ not ours	Go CDP driver, single binary deploys
g1879/DrissionPage	12,324	NOASSERTION	2026-07-22	⬜ not ours	Switch between HTTP mode and browser mode on the SAME session object
lightpanda-io/browser	33,439	AGPL-3.0	2026-08-04	⭐ starred	Purpose-built headless engine — ~10x lighter than Chrome for scraping
ultrafunkamsterdam/nodriver	4,613	AGPL-3.0	2026-05-13	⬜ not ours	Successor to undetected-chromedriver; no webdriver at all
ultrafunkamsterdam/undetected-chromedriver	12,786	GPL-3.0	2025-07-05	⬜ not ours	The classic detection bypass for Selenium
Kaliiiiiiiiii-Vinyzu/patchright-python	1,454	Apache-2.0	2026-07-16	⬜ not ours	Drop-in Playwright replacement, patched for stealth
daijro/camoufox	10,795	MPL-2.0	2026-08-03	⬜ not ours	Custom Firefox build with fingerprint spoofing compiled in
omkarcloud/botasaurus	5,626	MIT	2026-07-26	⬜ not ours	Scraping framework with anti-detect + caching + parallel built in
berstend/puppeteer-extra	7,390	MIT	2024-07-18	⬜ not ours	Plugin system (stealth, adblock) over Puppeteer
cdpdriver/zendriver	1,383	AGPL-3.0	2026-07-15	⬜ not ours	Async CDP driver, nodriver lineage

T2 — An LLM reads the page and decides the next action. Text-in, action-out. 14 tools 3 ours

Repo	Stars	Licence	Pushed	Ours?	What ONLY this one gives you
browser-use/browser-use	107,871	MIT	2026-08-04	⬜ not ours	The de-facto Python browser agent; DOM serialised for the model
browserbase/stagehand	23,728	MIT	2026-08-04	⬜ not ours	act/extract/observe primitives — mix natural language WITH Playwright code
Skyvern-AI/skyvern	22,671	AGPL-3.0	2026-08-04	⬜ not ours	Workflow-shaped: forms, 2FA, file downloads as first-class steps
lavague-ai/LaVague	6,385	Apache-2.0	2025-01-21	⬜ not ours	Compiles an objective into reusable Selenium code
vercel-labs/agent-browser	39,916	Apache-2.0	2026-08-04	⭐ starred	Agent-shaped CLI, lighter than a framework
steel-dev/steel-browser	7,420	Apache-2.0	2026-07-28	⬜ not ours	Open-source browser-as-an-API: sessions, proxies, captcha hooks
nottelabs/notte	1,987	NOASSERTION	2026-08-04	⬜ not ours	Turns a site into a structured action space for the model
microsoft/playwright-mcp	35,806	Apache-2.0	2026-08-04	⬜ not ours	Playwright as MCP tools — accessibility tree, not pixels
AgentDeskAI/browser-tools-mcp	7,283	MIT	2026-03-18	⬜ not ours	Reads console + network from YOUR live browser for debugging
executeautomation/mcp-playwright	5,626	MIT	2025-12-13	⬜ not ours	Playwright MCP server, broad tool surface
kontext-security/browser-use-mcp-server	825	MIT	2026-05-20	⬜ not ours	browser-use exposed over MCP
aeonfun/opendia	1,905	MIT	2026-08-04	⭐ starred	Drives the browser you're already logged into — no separate session
pinchtab/pinchtab	9,843	MIT	2026-08-03	⭐ starred	Multi-instance orchestration bridge — many browsers at once
MinorJerry/WebVoyager	1,113	Apache-2.0	2024-03-04	⬜ not ours	The reference eval-style web agent (research baseline)

T3 — Vision-grounded: screenshot in, x/y click out. Works with NO stable DOM. 13 tools 2 ours

Repo	Stars	Licence	Pushed	Ours?	What ONLY this one gives you
bytedance/UI-TARS	11,261	Apache-2.0	2026-01-27	⬜ not ours	The model itself — end-to-end GUI action. WE SERVE UI-TARS-2B
bytedance/UI-TARS-desktop	38,430	Apache-2.0	2026-08-02	💾 cloned	The agent shell around it, incl. browser + desktop operators
microsoft/OmniParser	25,222	CC-BY-4.0	2026-07-20	⬜ not ours	Turns a screenshot into labelled interactable elements for ANY vlm
web-infra-dev/midscene	14,489	MIT	2026-08-04	⭐ starred	Plain-English assertions on a UI; survives a redesign
OS-Copilot/OS-Atlas	453	Apache-2.0	2025-04-20	⬜ not ours	Open GUI grounding model + cross-platform action space
showlab/ShowUI	1,885	Apache-2.0	2026-04-24	⬜ not ours	Small vision-language-action model, light to run
zai-org/CogAgent	1,190	Apache-2.0	2025-04-04	⬜ not ours	GUI agent VLM from the GLM family
njucckevin/SeeClick	494	Apache-2.0	2025-07-13	⬜ not ours	The grounding baseline — click from instruction alone
X-PLUG/MobileAgent	9,038	MIT	2026-07-07	⬜ not ours	Mobile-first GUI agent family
simular-ai/Agent-S	12,120	Apache-2.0	2026-08-01	⬜ not ours	Agent that learns from its own experience across a GUI
microsoft/UFO	9,401	MIT	2026-07-08	⬜ not ours	Windows-native GUI agent across applications
xlang-ai/OSWorld	3,063	Apache-2.0	2026-07-28	⬜ not ours	The benchmark — how you PROVE a GUI agent works
web-arena-x/webarena	1,574	Apache-2.0	2025-11-26	⬜ not ours	Web-agent benchmark on self-hosted real sites

T4 — Whole computer, not just a browser. Most power, most brittle. 7 tools 2 ours

Repo	Stars	Licence	Pushed	Ours?	What ONLY this one gives you
openinterpreter/openinterpreter	67,563	Apache-2.0	2026-08-01	⬜ not ours	Natural language to code executed locally
bytebot-ai/bytebot	11,080	Apache-2.0	2025-09-12 `ARCHIVED`	⭐ starred 💾 cloned	Self-hosted containerised desktop the agent fully owns
e2b-dev/E2B	13,253	Apache-2.0	2026-08-04	⬜ not ours	Firecracker sandboxes for agent code execution
e2b-dev/desktop	1,441	Apache-2.0	2026-08-04	💾 cloned	Desktop sandbox with a stream you can watch
OpenAdaptAI/OpenAdapt	1,666	MIT	2026-08-03	⬜ not ours	Record a human doing it, then replay/generalise
m1k1o/neko	21,862	Apache-2.0	2026-08-04	⬜ not ours	Browser in a container with multi-user WebRTC view — human handoff
kasmtech/KasmVNC	5,159	GPL-2.0	2026-07-31	⬜ not ours	Streamed containerised desktop, web-native

A6 — Anti-bot survival. Without this, T1-T3 die in production. 7 tools 1 ours

Repo	Stars	Licence	Pushed	Ours?	What ONLY this one gives you
FlareSolverr/FlareSolverr	15,016	MIT	2026-07-16	⬜ not ours	Proxy that solves Cloudflare interstitials for your scraper
lexiforest/curl_cffi	6,230	MIT	2026-08-01	⬜ not ours	Python HTTP that impersonates a real browser's TLS+JA3
lwthiker/curl-impersonate	6,740	MIT	2024-07-18	⬜ not ours	The curl build behind TLS impersonation
bogdanfinn/tls-client	1,769	BSD-4-Clause	2026-07-02	⬜ not ours	Go TLS fingerprint spoofing
apify/fingerprint-suite	2,544	Apache-2.0	2026-08-01	⬜ not ours	Generate + inject consistent browser fingerprints
jhao104/proxy_pool	23,569	MIT	2026-06-15	⬜ not ours	Self-hosted rotating proxy pool
tiagozip/cap	7,432	NOASSERTION	2026-08-01	⭐ starred	Self-hosted CAPTCHA (for OUR forms, not for solving theirs)

A7 — Run many browsers at once, as infrastructure. 6 tools 0 ours

Repo	Stars	Licence	Pushed	Ours?	What ONLY this one gives you
browserless/browserless	13,553	NOASSERTION	2026-08-04	⬜ not ours	Browser-as-a-service: pooling, queueing, session limits
SeleniumHQ/docker-selenium	8,644	NOASSERTION	2026-08-04	⬜ not ours	Selenium Grid on Docker/K8s, the scale-out standard
aerokube/selenoid	2,655	Apache-2.0	2024-12-17 `ARCHIVED`	⬜ not ours	One container per session, far lighter than Grid
aerokube/moon	278	Apache-2.0	2026-04-03	⬜ not ours	Kubernetes-native browser cluster
novnc/noVNC	13,901	NOASSERTION	2026-06-06	⬜ not ours	Watch or take over a headless session from a browser
daytonaio/daytona	72,049	NONE	2026-07-24	⬜ not ours	Sandboxes for agent workloads, self-hostable

Re-prove any row: cd ~/ai-lab && python oss_lanes.py · python processes/bd_understand_business.py --report · ./run.sh && python score.py
Plan: SALES_AGENT_OS.md · architecture: AI_FACTORY.md · repo sparowjake-troll/getmax-ai-lab (private)