Everything here is plain markdown, readable on GitHub as-is. The same files
are the content source for the docs website in website/,
built with Next.js + Fumadocs:
cd website
npm install
npm run dev # local dev server with live reload
npm run build # static site in website/outThe site is deployed to GitHub Pages by
.github/workflows/docs.yml on every push to
main that touches docs/ or website/.
Conventions for writing pages:
- Every page carries a small YAML frontmatter block with its sidebar
title; keep the# H1in the body too — that's what renders, on GitHub and on the site. - Sidebar order and section groupings live in
meta.json(andposts/meta.json). - Keep writing ordinary relative links (
./other-page.md,../examples/...); the build rewrites in-docs links to site routes and out-of-docs links to GitHub URLs. - Write plain CommonMark, not MDX —
{and<in prose stay literal. This README andposts/harness-engineering-thread.mdare excluded from the site.
This directory mixes two audiences. Using Chidori is the path for agent authors and operators; Engineering notes are internal design records — useful history and rationale, but not tutorials, and some describe work that was completed, retired, or superseded.
Start here, roughly in order:
| Doc | What it covers |
|---|---|
| getting-started.md | Install, run the demos, inspect a durable run |
| your-first-agent.md | Tutorial: write an agent, pause it, replay it for $0, check it into CI |
| core-concepts.md | Host calls, the call log, safepoints |
| patterns.md | Task-oriented recipes: which primitive fits which job |
| faq.md | Python support, Node, providers, comparisons, data locality, troubleshooting |
| replay.md | Record, replay, resume, divergence rules |
| running-modes.md | run vs serve, policy profiles, --trusted |
| signals.md | Named signals: pause for humans or other agents |
| branching-execution.md | chidori.branch sub-runs |
| actors.md | Supervised, message-passing agent processes |
| detached-agents.md | Long-lived agents outside a session |
| context-management.md | Conversation and context windows |
| memory.md | chidori.memory: persistent cross-run key-value storage |
| template.md | chidori.template: Jinja prompt rendering |
| value-checkpoints.md | chidori.step: bounding replay cost |
| durable-storage.md | Run persistence, time travel (--until-seq) |
| package-management.md | Imports, node: builtins, npm packages |
| sandbox-model.md | The security model and its guarantees |
| observing-with-tael.md | OTLP export, run↔trace correlation, golden cases |
| deployment.md | Serving agents in production |
| Doc | What it covers |
|---|---|
| host-api.md | Every chidori.* method, option by option; providers; runtime policy |
| cli.md | Every subcommand and the approval postures |
Design records for contributors. Status headers inside each file are authoritative — several document retired or superseded work:
- architecture.md — engine + runtime layering
- conformance.md — Test262 methodology and CI gate
- captured-effects-vfs-crypto-timers.md — captured-effect surfaces
- interpreter-optimization.md — measured optimization phases
- js-performance-roadmap.md — profiling data and roadmap
- js-object-shapes-design.md — hidden-class design (implemented)
- jit.md — closure-threading JIT experiment (retired; kept as data)
- os-isolation-plan.md — process isolation design
- resume-performance.md — resume cost analysis
- dom-runtime-prototype.md — DOM runtime prototype
- ai-sdk-gap-analysis.md — feature comparison vs Vercel AI SDK
- consumer-usability-review.md — round 1: building a real agent on 3.6.0 (linear path)
- consumer-usability-review-2.md — round 2: the multi-agent surface (actors, branches, detached agents) under failure
- consumer-usability-review-3.md — round 3: the everyday-agent surface as a daily driver
- consumer-usability-review-4.md — round 4: the day-2 surface (npm packages, durable store, hydration, time travel,
verify) - consumer-usability-review-5.md — round 5: shipping to users (
servein production posture, SSE streaming, multiplayer signals under crashes, SDK-as-client, webhooks) - consumer-usability-review-6.md — round 6: the long-haul conversational surface (
init/chatfunnel, templates, cross-run memory, window compaction, local prompt cache) - branching-execution.md — also doubles as the branching design record
- rust-style-guide.md — contributor conventions
- releasing.md — release train and versioning