This repo is the curated developer documentation assembly for building on Autonomi and is intended to power what developers consume through docs.autonomi.com/developers.
The upstream repositories remain the implementation truth. This repo brings their merged surfaces together into one coherent developer experience so people do not have to jump between multiple repos and reconstruct how the network and tooling fit together.
It is designed to serve several purposes at once:
- a GitBook-driven, human-navigable documentation space alongside the rest of the Autonomi docs
- structured Markdown that remains useful for AI agents and LLM retrieval workflows
- AI-agent-friendly, parseable content designed for
llms.txt-style consumption and GEO-aware discovery - a verification layer that ties published claims back to specific upstream commits
- a future automation target that can detect upstream drift and keep the docs aligned
This repo currently focuses on the Autonomi developer documentation experience.
It documents four primary interfaces:
- Build with the SDKs
- Use MCP with AI tools
- Use the
antCLI - Build directly in Rust
The documentation is written primarily for human developers encountering Autonomi for the first time, while also remaining parseable and useful for AI agents.
- It is not the only source of truth for technical behavior. The upstream repos are.
- It is not a copy of every upstream README.
- It is not intended to describe unmerged branches or future-state product plans.
The repo retains two verification mode names:
target-manifestcurrent-merged-truth
This is the authoritative mode for public technical documentation.
target-manifest.yml records reviewed public releases and their exact source and distribution evidence. New or refreshed public claims use these pins instead of moving default branches. If a component has no reviewed release profile, its technical update is deferred until that evidence exists.
Existing blocks in this mode are historical migration debt. They remain inspectable but cannot be advanced or used to support a new or refreshed public claim.
Every documentation task follows this workflow:
- Source audit
- Draft
- Verify
That workflow is defined in:
CLAUDE.mdplanning/verification-workflow.md
Determine:
- page type
- relevant source repos
- exact refs and SHAs
- canonical source artifacts
Write only from audited evidence.
Check every technical claim, endpoint, command, type, field, and example against the audited sources, then record real verification metadata in the page.
Verified pages include HTML comment blocks like this:
<!-- verification:
source_repo: ant-sdk
source_ref: [release tag]
source_commit: [release commit SHA]
verified_date: [YYYY-MM-DD]
verification_mode: target-manifest
-->Rules:
source_commitmust be a real commit SHA- multi-repo pages must use multiple verification blocks
- historical notes and prior synthesis docs are not authoritative
- baseline changes move atomically with
target-manifest.ymland a dated promotion record - legacy
current-merged-truthblocks stay byte-identical until their claims are migrated
repository-root/
├── .gitbook.yaml # GitBook config; published content lives in docs/
├── CLAUDE.md # Repo rules for AI-assisted docs work
├── repo-registry.yml # In-scope source repos and source-of-truth config
├── component-registry.yml # Maps product components to docs pages
├── docs/ # Published documentation content
├── planning/ # Workflow docs, plans, and working process notes
└── reference/ # Supporting reference material and research notes
Important:
- Only
docs/is intended to be published via GitBook planning/,reference/, andCLAUDE.mdare working materials, not published docs content- GitBook should sync the repo root and use
.gitbook.yamlto scope published content todocs/
This repo is intended to be reviewed in two places.
Use GitHub review for:
- technical accuracy
- source alignment
- wording and structure
- verification metadata and SHAs
- repo/process changes
Use GitBook review for:
- navigation
- page flow
- heading readability
- tabs, tables, and code block rendering
- overall user experience
A page is publishable when:
- it has been source-audited
- it has been verified against real upstream SHAs
- any required runnable workflow checks have passed
- it renders cleanly in GitBook
- it is ready for an external developer audience
This repo is designed to become progressively more automated.
The intended model is:
- observe public releases for enrolled components
- reconcile public artifacts with exact source identities
- map release candidates to affected docs pages
- re-run source audit, draft, and verification
- promote the manifest and verification blocks only after human review
The files that support that model are:
repo-registry.ymlcomponent-registry.ymlplanning/verification-workflow.md
At the moment, the repo is still in an agent-assisted hardening phase. Automation is being introduced deliberately rather than all at once.
Current working model:
- GitHub repo for content review and version control
- GitBook sync for unpublished preview and rendering review
Planned direction:
- finalize the verified baseline
- review with developers from GitHub and GitBook preview
- introduce upstream-triggered automation in this repo
If you are editing or reviewing docs here:
- read
CLAUDE.md - use
planning/verification-workflow.md - treat reviewed release profiles in
target-manifest.ymlas the authority for public claims - keep internal provenance language out of rendered docs prose
- favor developer-facing paths and tasks over repo-shaped framing
Key writing expectations include:
- explain what a tool is before telling the reader to install or run it
- prefer user outcomes and choices in titles
- avoid repo names as primary page titles unless the repo itself is the topic
- keep headings readable in GitBook
- use cURL, Python, Node.js / TypeScript, and Rust as the default featured SDK examples when supported and verified
See repo-registry.yml for the authoritative in-scope repo list.
See component-registry.yml for which components and repos inform each page.