Documentation that knows what it knows.
AgentDoc is a compiler for organizational knowledge. Write readable source files. Get typed, evidence-backed objects with lifecycle, ownership, and scope — for humans and the agents acting on their behalf.
$ brew install adoc- status
- verified
- owner
- @platform-billing
- scope
- env=prod · plan=*
- evidence
- 2 sources · current
- verified
- 2026-05-02 · expires 88d
Markdown says things.
It does not know them.
A paragraph cannot tell you who wrote it, when it was last verified, what evidence it stands on, or whether an agent is allowed to act on it. AgentDoc encodes all of that — without taking the readability away.
Credits are deducted after generation completes.
- Is it still true?
- Who wrote it, who owns it?
- What evidence supports it?
- Does it apply to staging or only prod?
- Was it verified by tests, code, or just a meeting?
- May an agent act on it?
“Credits are decremented after generation completes successfully.”
- Verified · expires in 88 days
- Owned by @platform-billing
- 2 evidence sources, currently passing
- Scope: env=prod, plan=*
- Linked to credits.rs#L142
- Agents: read · cite · not patch
Eight types. One graph.
Every sentence in your docs becomes an object of a specific kind. Each kind has its own schema, lifecycle rules, and agent permissions. The compiler enforces them.
Agents do not follow prose.
They follow objects with owners,
evidence, and a lifecycle.
— AgentDoc thesis · v0.1
Wherever you write,
wherever you ask.
The same compiler powers a CLI, an IDE language server, a query API, and a web app. One source of truth. Many ways in.
A compiler that fails loudly.
adoc checkvalidates schemas, freshness, scope coherence, and contradictions. Errors point to source lines with diagnostic codes — not “something is off”. CI-ready, single-binary, no daemon.
- Diagnostic codes (E0341, W0118) you can grep
- Stale evidence detection on every build
- Strict mode for regulated knowledge
- Output as
docs.html+docs.agent.json
“Credits are decremented after generation completes successfully, except during pending reconciliation.”
A workspace for organizational memory.
Browse claims, decisions, and constraints by owner, scope, and lifecycle. Open any object to see its evidence, its history, its agent permissions, and the patches waiting on a reviewer.
- Evidence inspector with file + line links
- Semantic diff: not what changed, what it means
- Contradiction inbox with severity
- Per-object agent permission panel
Diagnostics where you already are.
The AgentDoc LSP runs in your editor. It checks every keystroke against the graph: stale claims, missing owners, contradictions, scope drift. Quick-fixes write valid metadata for you.
- Inline diagnostics with code citations
- Hover-peek for referenced claims
- Quick-fix code actions for E0341, W0118, I0042
- Workspace-wide find-references on any object ID
What a knowledge graph
does for you.
Beyond compilation: an evidence layer, a semantic diff engine, an agent patch protocol, and a permission model — all driven by the same source files engineers already write.
.adoc is plaintext.Boring infrastructure,
shipped quietly.
A compiler today. A graph next. A web app and IDE after that. Enterprise governance and an agent registry at v1.0. We ship in increments because correctness is a feature.
- adoc check / build
- 11 object kinds
- Diagnostic codes E0001–E0399
- adoc query
- graph index format
- relation walks
- app.agentdoc.dev
- adoc-lsp
- code actions / quick fixes
- agent registry
- patch proof obligations
- SOC 2 export
Write what you believe.
Compile what you know.
The v0.1 compiler is open source today. Run it on a folder of .adocfiles. See your organization's beliefs typed for the first time.