PR Notes,

im not taking pr right now, wait for beta version, for now use the version that works for you 1.0.5, 1.0.6, etc.

Cartethyia

Cartethyia is a Bun + Elysia AI proxy with a public landing page and an authenticated console. It translates OpenAI Chat/Responses and Anthropic Messages requests while routing models across provider accounts, aliases, combos, and custom compatible endpoints.

Current release: 1.0.7-alpha

Community

[Cartethyia Home Discord ]

Community access is free. Join Discord: https://discord.gg/zFcNPJM6qM

ShowCase

Features

• OpenAI Chat Completions, Responses, and Anthropic Messages.
• Provider routing with priority, round-robin, cooldowns, aliases, combos, and failover.
• Provider accounts, API keys, model ACLs, usage limits, logs, and request history.
• Model Studio with persistent history, edit/copy/delete actions, token usage, and compaction.
• Custom OpenAI-compatible and Anthropic-compatible upstreams.
• Responsive Cartethyia public landing page at /.
• OAuth-backed provider accounts for Codex, Anthropic, Grok CLI, Google Antigravity, and Kiro, including Kiro device authorization and token import.
• Kiro model routing through kiro/<model> with AWS EventStream decoding; quota checks use a 15-minute cooldown to avoid unnecessary upstream traffic.
• Customization at /console/customization: frosted-glass custom background, optimized seasonal lock effects, custom seasonal images, frequency, and size controls.
• Public API-key share pages with usage, budget, allowed-model visibility, copy controls, and an explicit disabled-key state.

Quick start

bun install
cd dashboard && bun install && cd ..
cp .env.example .env
bun run dev

Open:

• Public page: http://localhost:12800/
• Console: http://localhost:12800/console/
• Health: http://localhost:12800/health

API

Route	Protocol
POST /v1/chat/completions	OpenAI Chat Completions
POST /v1/responses	OpenAI Responses
POST /v1/messages	Anthropic Messages
GET /v1/models	Unified provider/model catalog
GET /health	Liveness probe

Create a proxy API key from Console → API Keys. Keys can restrict providers/models and enforce request, concurrency, and token limits. Share links remain useful when a key is disabled: visitors see the key status and an explanation instead of an empty placeholder page.

curl http://localhost:12800/v1/models \
  -H "authorization: Bearer $CARTETHYIA_API_KEY"

Configuration

Copy .env.example for local development. In production, set secrets in the platform and mount DATA_DIR as persistent storage.

Variable	Purpose
PORT	HTTP listener; Railway supplies this automatically.
DATA_DIR	Persistent configuration, logs, and runtime state.
CONSOLE_PASSWORD	Console login password.
CONSOLE_JWT_SECRET	Secret used to sign console sessions.
PROXY_AUTH_MODE	open or api_key.
CONSOLE_SESSION_TTL_HOURS	Console session lifetime; defaults to 12.
TRACK_PAYLOADS	Request/response tracking level: none or meta.
TRACK_ASSETS	Asset tracking level: none, meta, or store.

Docker / Railway

docker build -t cartethyia .
docker run --rm -p 12800:8080 \
  -e PORT=8080 \
  -e DATA_DIR=/app/data \
  -e CONSOLE_PASSWORD=change-me \
  -e CONSOLE_JWT_SECRET=replace-with-a-long-random-secret \
  -v cartethyia-data:/app/data \
  cartethyia

For Railway, mount a volume at /app/data, configure CONSOLE_PASSWORD, CONSOLE_JWT_SECRET, and TRUST_PROXY=true, then verify GET /health after deployment.

Development

bunx tsc --noEmit -p .
bun test test/
cd dashboard && bun run test && bun run build

See the protocol notes in docs/, the release notes in docs/commit-notes/1.0.7.md, and the landing mockup in docs/landing-page-mockup.md. Contributions should follow CODE_OF_CONDUCT.md and the pull-request template.