Self-hosted open-source AI gateway routing LLM traffic at 0.5 ms overhead, with budgets and guardrails.
Homepage hero
1 / 3Overview
What is Antirion?
Your AI gateway should add 0.5 ms, not 20
Antirion is an open-source, self-hosted AI gateway for developers. It is a single Rust binary with the admin console built in, sitting between your applications and every LLM provider. Point any OpenAI or Anthropic SDK at it and get multi-provider fallbacks, per-team and per-key budgets, PII and secrets redaction, caching, and a log of every request with its route and cost. It claims 0.5 ms added latency at 5,000 RPS on 4 vCPUs. Key capabilities: fallback chains and weighted, least-cost or least-latency routing across OpenAI, Anthropic, Gemini and Bedrock; dollar or token budgets per organization, team, project, key and model; edge guardrails (PII/secrets redaction via regex, JSON schema, tool permissions, moderation and LLM judges) on requests, responses and streams; full observability with Prometheus, OTLP and log sinks; exact and semantic caching; SSO/SCIM; MCP and A2A agent gateway support; deployable as binary, Docker, Compose or Kubernetes Helm chart. Licensed Apache 2.0, free with all features included.
Platforms and languages
Antirion Availability
Platforms
Languages
Capabilities
Antirion Key Features
Sub-millisecond routing
Single Rust binary gateway adding about 0.5 ms per request at 5,000 RPS, with built-in admin console included.
Multi-provider fallback chains
Fallback chains, weighted splits, least-cost and least-latency routing across OpenAI, Anthropic, Gemini and Bedrock, with retries and health checks.
Spend budgets
Dollar or token budgets per organization, team, project, key and model to keep LLM spend under control.
Edge safety guardrails
PII and secrets redaction via regex, JSON schema, tool permissions, moderation and LLM judges, on requests, responses and streams.
Full request observability
Logs every request's route, attempts, failures, tokens, cost, cache hits and guardrail verdicts, with Prometheus, OTLP and log sinks.
Easy deployment
Runs as a single binary, Docker container, Docker Compose cluster or Kubernetes Helm chart; supports SSO, SCIM, clustering, exact and semantic caching.
Best for
Who uses Antirion?
Platform engineering teams and backend or AI developers
Unify access to multiple LLM providers behind one fast API with low latency, budgets and guardrails
Plans and access
Antirion Pricing
Free
Free and open source under Apache 2.0; all features included, no paid tier.
Common questions
Antirion FAQs
What is Antirion?
An open-source AI gateway: one Rust binary with a built-in admin console that sits between your apps and every model provider, routing, retrying, budgeting, guarding and recording every LLM call behind a single API.
How much does it cost?
Nothing. Antirion is open source under the Apache 2.0 license with every feature included: SSO, SCIM, the audit log, clustering and guardrails are not held back for a paid tier.
Will it work with my existing stack?
Clients speak the OpenAI, Anthropic, Gemini or Bedrock APIs, so official SDKs, LangChain and the Vercel AI SDK work by just changing the base URL. The gateway translates between provider dialects.
Where does my data go?
Nowhere new. Antirion runs in your own infrastructure and requests go straight from your network to providers. Request content is not logged unless a policy enables it, and provider keys are sealed at rest.
How fast is it really?
The project reports adding about 0.5 ms at 5,000 RPS on 4 vCPUs, holding 14,000 RPS. The benchmark harness (vs Bifrost and LiteLLM on dedicated AWS machines) is published in the repository.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.
