What it replaces
Stop spreading model logic across every product team.
Product leaders get one place to understand readiness, risk, and usage. Engineering teams get a stable OpenAI-compatible surface that does not change every time the provider stack changes.
Use presets for Ollama, vLLM, Bedrock, Hugging Face, OpenRouter, OpenAI, and more.
Segment models by chat, embeddings, OCR, vision, video, code, local, or production use.
Draft route changes, simulate fallback order, and promote only after review.
See request, response, guardrail decisions, redaction type, and original access audit.
Provider catalog
Connect the AI stack you already have, then add the stack you want next.
OpenAI
Anthropic
Bedrock
Hugging Face
OpenRouter
vLLM
Ollama
Groq
Mistral
Together
For product teams
Know what is safe to launch
Readiness, route drafts, incidents, and guardrail coverage are visible without reading code.
For platform teams
Govern once, reuse everywhere
Projects, environments, service keys, and policy sets keep every app on the same operating model.
For developers
Keep the integration simple
Call one OpenAI-compatible endpoint and let the gateway handle provider choice, fallback, and logs.
For security
See what changed and why
Guardrail activity, payload diffs, redaction reasons, and original-payload access live in Logs.
Open-source core
Everything needed to run a real AI gateway.
- Provider registry and OpenAI-compatible gateway API
- Model discovery for supported provider catalogs
- Routing, fallback, virtual keys, logs, and local runtime support
- Developer-friendly setup scripts and Docker workflow
Enterprise controls
Add commercial controls when the deployment grows.
- SSO, SCIM, workspace governance, and richer tenant controls
- Route approvals, policy coverage, Splunk export, and support bundles
- Production posture, readiness reports, and operational diagnostics
- License-based upgrade path without changing the core gateway API
Developer documentation
Technical enough for builders. Clear enough for buyers.
Keep the homepage customer-friendly, but give engineers a direct path to the details they need before they trust a gateway in front of production AI traffic.
API
OpenAI-compatible gateway
Base URL, auth headers, chat, embeddings, streaming, and OpenWebUI setup.
Models Catalog and model rolesDiscover provider models and classify them as chat, OCR, vision, video, code, or embeddings.
Safety Guardrails and policy setsSecrets, PII, PHI, harmful content, denial advice, jailbreaks, and response sanitization.
Ops Logs as source of truthRequest/response view, diff, guardrail reason, original payload access, and audit timeline.
Deploy Docker, AWS, and secretsRun locally, push images to ECR, deploy to ECS, and manage database and license secrets.
Providers Connect and scan runtimesUse presets, store credentials, scan models, and expose only approved model targets.
Developer handoff
One integration path from prototype to production.
Teams can start with a local provider, move to hosted inference, and keep the same app integration while platform owners control routes, keys, guardrails, and audit trails.
OpenAI-compatible request
curl https://gateway.example.com/v1/chat/completions \
-H "Authorization: Bearer sk_live_..." \
-H "Content-Type: application/json" \
-d '{
"model": "chat",
"messages": [{ "role": "user", "content": "Summarize this contract" }]
}'Start small, keep the path to enterprise open