CompanyAgentWiki

A reusable second brain for a whole company that your AI agents can read. It turns every knowledge source — source code, Confluence, Jira, and legacy documents — into one linked, semantically-searchable Markdown vault, indexed by HumanAgentWiki (a local PostgreSQL + pgvector Markdown indexer, exposed over MCP).

The one rule: HumanAgentWiki indexes Markdown only. So every source gets a small pipeline that produces Markdown into this vault; the database is then a disposable mirror of those files. Full design: SECOND_BRAIN_DESIGN.md.

Layout

The top-level folder of each file becomes its graph category, so the layout is deliberate:

Quick start

Using Claude Code? Just ask for the **install-guide** agent (.claude/agents/install-guide.md) — it runs preflight checks, asks where your vault lives and which repos to map, edits the manifest, runs the installer, generates the docs, and stands up the brain, confirming before anything slow. The manual steps it automates:

Full step-by-step for a fresh machine: see DEPLOY.md.

Vault vs. toolkit. This repo is the engine. The vault (where docs are written and what HumanAgentWiki indexes) is chosen at install time and recorded in brain.config.json — so one clone of the toolkit can drive a separate hub repo as its brain. VAULT_DIR overrides it per-run.