Agnos LLM gateway
Contain LiteLLM, Bifrost, Portkey.. behind one control plane
3 followers
Contain LiteLLM, Bifrost, Portkey.. behind one control plane
3 followers
Agnos is an OpenAI-compatible control plane that wraps the LLM gateway you already run. Keys stay in your encrypted vault (the gateway gets one per request and stores none); you get budgets, guardrails, routing, cost attribution and audit, and you can swap or combine gateways (litellm, bifrost, portkey etc) with a config change. MIT, self-hosted. Live demo, https://agnos-llm-gateway.site https://github.com/siva010928/agnos-proxy-oss
Interactive
Free
Launch Team / Built With
World model AI : Sharm SEKAISimulate your idea on a living world.
Promoted
We open-sourced Agnos Proxy: a control plane that contains the LLM gateway you already run.
Most gateways fuse two jobs - translating your request to each provider, and governing it (keys, budgets, guardrails, cost, audit). Fuse them and your provider keys live inside the code parsing untrusted input, and switching gateways means a rewrite.
Agnos separates them. It contains LiteLLM, Bifrost, Portkey, or a built-in engine as swappable adapters behind one OpenAI-compatible endpoint:
Change one line, keep your SDK.
Route each provider to the gateway that serves it best, or swap one at runtime.
Keys stay Fernet-encrypted in your vault, injected per request. The gateway stores nothing.
Budgets that stop spend before it happens, PII/secret guardrails, cost attribution and audit, the same across every gateway.
Self-hosted, MIT, about 1 ms overhead.
Live playground: agnos-llm-gateway.site
Repo: github.com/siva010928/agnos-proxy-oss
Feedback and "which gateway should we contain next" welcome.