Company Intelligence - starter template

A single-source-of-truth Git repo of plain Markdown that serves three audiences at once:

• Humans read it in their editor, GitHub, Obsidian, or a rendered MkDocs site.
• AI agents (Claude Code, Cursor, Codex, OpenClaw, Gemini, Windsurf, Aider…) read it via AGENTS.md.
• Live systems (Linear, Slack, Notion, Granola, Google Workspace) plug in through MCP.

Durable truth lives here. Execution state stays in the systems that own it. Agents synthesize across both.

📋 Need a quick fact or an official link (app/store URL, socials, company no. / EIN / DUNS, founders, funding), or filling an application/form/deck? See FACTSHEET.md, the single dense quick-reference. It ships as a placeholder template; fill it with real values during setup (Step 3 below).

How it fits together

flowchart LR
  subgraph Sources["Raw sources (high volume, low signal)"]
    M[meetings/ transcripts]
    I[inbox/extracted/ staged facts]
  end
  subgraph Brain["Compiled truth (what agents and humans read)"]
    R[company.md strategy.md product.md brand.md ops.md finance.md]
    E[customers/ competitors/ people/ sales/ suppliers/ distributors/]
    T[decisions/ weekly/ FACTSHEET.md]
  end
  subgraph Agents["Every agent, one contract"]
    A[AGENTS.md + folder READMEs]
    C[Claude Code / Cursor / Codex / OpenClaw]
  end
  M --> I -->|curation contract: SAFE facts auto-land, SENSITIVE facts wait for a PR| Brain
  A --> C
  C -->|reads| Brain
  C -->|writes via PR| Brain
  L[Live systems via MCP: Linear, Slack, Notion, Calendar] -.->|freshness, never long-term memory| C

Loading

What it looks like in use

you:    What did we promise Acme in the last renewal call, and who owns the follow-up?
agent:  From customers/acme-co.md (updated 2026-05-21, last_verified 2026-05-21):
        - Renewal at the current tier until 2027-03; a 20% volume discount was offered
          if they add the EU entity (decisions/2026-05-21-acme-eu-discount.md).
        - Open follow-up: security questionnaire owed by us, owner Sarah, due 2026-06-01
          (meetings/2026-05-21-acme-renewal.md, action items).
        Nothing newer is on record; the next Acme touchpoint is not scheduled yet.

Every claim points at a file, every file carries frontmatter (type, status, owner, updated, last_verified) that CI validates, and nothing lands on main without the curation contract in docs/CURATION.md.

What's "intelligence"?

The intelligence is the persistent knowledge layer your company runs on: an operating layer, not just storage. Agents and humans both read from it. Both write to it. PRs gate the truth.

Who this is for

• Founders who want their AI agents to actually know their company
• Small teams (1-20 people) who don't need enterprise knowledge graphs
• Consultancies and agencies who want a reusable client template
• Anyone tired of repeating company context to every new chat

What you get

<company>-intelligence/
├── Entry files          AGENTS.md, README.md, INDEX.md
├── Quick reference      FACTSHEET.md (facts + official links, for applications/forms/decks)
├── Root knowledge       company.md, strategy.md, product.md, brand.md, ops.md, finance.md
├── Entity folders       customers/, competitors/, people/, sales/
├── Time-series          meetings/, decisions/, weekly/
├── Agent infrastructure skills/ (optional), templates/
└── Config               .gitignore, .github/workflows/

What to sync into an AI (context vs raw sources)

This repo is compiled truth + raw sources. The root knowledge files and entity folders already distill every meeting and document - so when you sync the repo into Claude project knowledge (or any AI) for company context, include the compiled brain and skip the raw sources, which are redundant and the capacity hogs.

The calls are not summarised file-by-file - but their substance is already in the compiled pages. The compiled pages are the summary of every call.

Tier	Paths	Notes
✅ Always sync (the compiled brain)	FACTSHEET.md; company.md · strategy.md · product.md · brand.md · ops.md · finance.md; customers/ · competitors/ · people/ · sales/; curated decisions/; weekly/; INDEX.md · README.md	Small, pure signal: this alone is full company context. The fact sheet is tiny and highest-signal; include it first.
🔶 Optional (only if you have headroom)	decisions/imported/ (if present - granular auto-extracts) · the last 2-4 strategic meetings/ (leadership/product/model)	All text/cheap; adds detail but lower-signal than the compiled pages.
❌ Don't sync	meetings/ bulk - especially standups · any chart images · .github/ · scripts/ · skills/ · templates/	Content already compiled above; images + transcripts are the capacity hogs; the rest is machinery.

Rule of thumb: text = high signal per token, images = low. Even with headroom, low-value files dilute what the AI retrieves. Full rationale: docs/CLAUDE_SYNC.md.

Quick start (≤30 minutes)

See docs/SETUP.md. It walks you from empty clone to working intelligence with seeded strategy, one skill installed, and Claude Code reading the repo correctly.

For agents starting fresh

If you're an AI agent reading this repo for the first time - your operating contract is AGENTS.md. Read it before doing anything else.

If you've been asked to build an intelligence repo for a company from raw, unstructured data, read these two, in order:

1. docs/DATA-ORGANIZATION-PLAYBOOK.md - what goes where and why, plus how to fill the six root files. The routing logic.
2. docs/AGENT-INSTRUCTIONS.md - the full nine-phase migration procedure with human checkpoints.

Then each folder's own README.md is the complete file-generation contract for that folder (frontmatter, fields, sourcing, worked example, quality bar, edge cases). An agent with no prior context can generate correct, consistent files from those READMEs alone.

How to use this as a template

1. Click Use this template on GitHub (or gh repo create --template)
2. Name the new repo <company>-intelligence
3. Make it private
4. Clone locally, run find/replace on {{COMPANY_NAME}}, {{COMPANY_SHORT}}, {{FOUNDER_NAME}}, etc.
5. Follow docs/SETUP.md

For founders applying this to their own company: budget one afternoon for setup, one week for seed content. For consultancies applying it to a client: budget a 2-hour kickoff workshop plus one week of async ingestion.

Design principles (the one-paragraph version)

One durable truth (Git) + many live sources (MCP). Prefer file-native context before retrieval systems. Separate stable rules from fluid facts. Keep always-loaded files small. Put repeatable know-how in Skills, not in the root prompt. Use subagents for isolation. Plan, then act, then compact. Use MCP for freshness and narrow writes, not as long-term memory. Assume prompt injection is real. Don't add RAG, vector DBs, or graph DBs until your traces prove you need them.

See docs/ARCHITECTURE.md for the full rationale.

License

MIT for the template structure (file layouts, contracts, scripts, documentation) - see LICENSE. Content you create in a repo instantiated from this template is yours and is not subject to this license.

---

Maintained by Agentmatik. Based on consensus across independent deep research on agent-ready company context patterns (Anthropic, Cline, gbrain, Superpowers, Chroma context-rot research, LangChain context engineering, AGENTS.md standard).

Using this brain with gbrain

This repo ships a brain-resident gbrain skillpack (skillpack.json, skills/company-brain/). A connecting harness discovers it on gbrain sources add and reads skills/company-brain/SKILL.md.

Sync it with the company-brain schema pack or gbrain retypes customer, decision, strategy and the rest of this layout to note. The skill has the commands.