Adapted from the black-candle-technologies/chitin repository docs.
Cloud-hosted AI gateway that reduces LLM token costs 40%+ via caching, semantic dedup, context compression, and intelligent model routing. Clients point any OpenAI-compatible SDK at the gateway; no infrastructure changes required.
| Path | Description |
|---|---|
apps/app |
Next.js dashboard (self-serve tenant UI) |
apps/web |
Marketing website (Next.js) |
services/backend |
Go API gateway (the core proxy + pipeline) |
sdks/ts |
TypeScript client SDK (@chitin/sdk) |
sdks/python |
Python client SDK (chitin) |
packages/eslint-config |
Shared ESLint configuration |
scripts/ |
Repository-level automation scripts |
pnpm install # all JS/TS dependencies from the repo root
Then see the area docs:
docs/services/backend/README.md → Backenddocs/apps/app/README.mdEach workspace contains an AGENTS.md with coding-agent guidance specific to that area. Read the relevant AGENTS.md before working in any sub-directory; the root AGENTS.md has the index.
The repo's docs/INDEX.md organizes everything by area:
Backend (services/backend): quickstart, API reference, optimization guide, billing/Stripe, Anthropic savings report, dashboard implementation guide, operations runbook, plus specs (product requirements, engineering roadmap, admin auth spec, compression guide, competitive analysis, enterprise migration).
Dashboard (apps/app): product spec (SPED), tech-stack rationale, ADR index (including ADR 0001 Next.js BFF).
SDKs: cross-language SDK spec, per-language READMEs, changelogs, roadmaps, API references, CLI reference.
~42,700 lines across 317 files: 84.7% Go, 8.5% TypeScript, 6.7% Python.