DOCUMENTATION

Build with Routiqel.

Reference for the local gateway MVP. Features that are planned or unavailable are identified in context.

Introduction

Routiqel is a developer-first control plane for AI applications. The local MVP provides a non-streaming OpenAI-compatible endpoint and a deterministic adapter.

The gateway currently supports chat completions with model, messages, temperature, and max_tokens. It records request metadata and estimated cost without saving prompt text by default.

Quickstart

Run the Routiqel frontend and API locally, then create a project API key and call the sandbox endpoint.

The public website does not host accounts or a gateway. Follow the repository README to start the frontend and API on your machine, create a workspace at /app/signup in that local environment, and copy an API key from the local dashboard. The secret is only shown once.

curl http://localhost:8000/v1/chat/completions \
  -H "Authorization: Bearer rtq_YOUR_PROJECT_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"routiqel-demo","messages":[{"role":"user","content":"Hello"}]}'

Authentication

Dashboard requests use an HttpOnly session cookie. Gateway requests use a project-scoped bearer API key.

Passwords are hashed before storage. API key secrets are returned only during creation; the database stores a hash. Use HTTPS and a production secret manager before exposing a deployment.

API keys

Create and revoke project-scoped keys from the dashboard.

Keep keys on trusted servers. Never expose a gateway key in browser JavaScript or commit it to source control. Revoked keys stop authenticating immediately.

OpenAI-compatible API

POST /v1/chat/completions accepts a documented subset of the OpenAI chat completion request shape.

Supported request fields: model, messages, temperature, and max_tokens. Streaming is not supported. The deterministic demo adapter returns simulated text and estimated token usage.

Provider configuration

The local MVP supports the deterministic demo adapter and optional connections to fixed OpenAI-compatible provider endpoints.

Provider credentials are stored encrypted by the API server and are never returned to the browser. Arbitrary base URLs are not accepted. Provider calls may incur charges from the account owner.

Routing

Choose a primary provider, optional demo fallback, retry count, and timeout in the Routiqel app.

Retries are bounded and only used for supported transient failures. The MVP does not implement complex policy routing or latency-based model selection.

Request logs

Request logs contain trace ID, timestamp, model, provider, status, latency, token counts, and estimated cost.

Prompt and completion content are not stored by default. Logs are persisted locally in SQLite and are scoped to the project that authenticated the request.

Error codes

The gateway returns standard HTTP errors with a safe JSON detail message and trace ID.

Common responses include 401 for invalid project keys, 402 when a configured hard budget is reached, 422 for invalid payloads, 429 for rate limits, and 502/504 for upstream failures.

Rate limits

Basic in-process rate limits protect local authentication and gateway endpoints.

Limits reset when the API process restarts and are not a substitute for a shared production rate-limit store or abuse controls at the edge.

Pricing and usage

Routiqel has no billing integration or paid plans in this MVP.

Cost values are estimates based on recorded token counts and demo rates. They are not provider invoices and should not be used as accounting records.

Security

The local MVP uses project-scoped authorization, hashed API keys, encrypted provider credentials, validation, and no prompt logging by default.

Before production deployment, review HTTPS, secret rotation, session storage, shared rate limits, backups, tenant-isolation testing, deletion and retention controls, and managed key storage. No SOC 2, ISO 27001, HIPAA, or other certification is claimed.