CogniGate gateway: self-hosted, multi-tenant, OpenAI-compatible LLM gateway.
1.5K
The request path of CogniGate, a self-hosted, multi-tenant LLM gateway. Applications hold one CogniGate key and speak the OpenAI API; provider credentials stay inside the deployment.
It handles authentication, capability aliases (fast, balanced, best), fallback chains, circuit breakers, rate limits and quotas, and optional response caching. Usage is sent to the companion cognigate-analytics image for durable storage.
latest: the newest build of mainX.Y.Z and X.Y: releaseslinux/amd64 and linux/arm64, with build provenance attestations. The same images are on ghcr.io/life-experimentalist/cognigate-gateway.The supported way to run CogniGate is the full stack (gateway, analytics, Postgres):
curl -sSL https://cognigate.vkrishna04.me/install.sh | bash
Read the script first if you prefer; it clones the repository, generates credentials and starts docker compose.
The gateway on its own:
docker run -p 8080:8080 \
-e CG_ADMIN_BOOTSTRAP_KEY=<generate with: openssl rand -hex 24> \
vkrishna04/cognigate-gateway:latest
Without ANALYTICS_URL and ANALYTICS_TOKEN the gateway keeps usage in memory, and a restart loses it.
| Setting | Purpose |
|---|---|
port 8080 | OpenAI-compatible API, admin API, /healthz |
CG_ADMIN_BOOTSTRAP_KEY | first admin credential, used to create tenants and keys |
ANALYTICS_URL, ANALYTICS_TOKEN | where usage is written |
CG_QUOTA_ENFORCEMENT | on (default) or observe |
CG_CACHE_ENABLED | true to turn on response caching |
-config <path> | optional configuration file; mount it and pass the flag |
Content type
Image
Digest
sha256:babec14d5…
Size
11.1 MB
Last updated
11 days ago
docker pull vkrishna04/cognigate-gateway