Awesome Infra for AI › LLM Gateways & Proxies

MaySudo/Misceo

⭐ 403 added to this list on 2026-08-31 repository created 2026-07-18

Misceo is a local gateway that lowers the cost of running AI agents without breaking their conversations or tool loops. It exposes an Anthropic-compatible endpoint, so existing clients point at it instead of the provider, and it decides per request which backend should serve the answer. Unlike a static router that must commit before generation, Misceo can also inspect the first completed candidate and then choose what the client receives. A request flows through explicit precedence rules first: protected routes, open tool loops, active pins and user rules bypass the cascade. Eligible traffic goes to a configured lower-cost backend. A structural gate rejects upstream failures, empty output and malformed tool calls. Depending on the selected mode an optional judge scores the visible candidate against the latest user turn, and a rejected candidate is discarded while a stronger backend regenerates the served answer. Escalated conversations then stay on the stronger backend for a configurable number of turns. The project puts particular weight on agent state. Requests are grouped by structured session identity so bootstrap, title-generation and visible turns from one launch stay in the same conversation. At model-family boundaries it handles provider-specific thinking signatures, cache markers, model names and message invariants so a handoff does not corrupt the transcript. When a model opens a tool loop, that backend keeps ownership until the loop completes, and the first visible reply of a session is served by the strong backend. Three postures ship out of the box: quality-first, balanced and savings-first. The proxy, an embedded dashboard, the configuration and the traffic logs all run locally, though inference itself still reaches the configured providers. Misceo is distributed as prebuilt binaries through npm under FSL-1.1-ALv2; this repository holds documentation, issues and releases rather than source.

https://github.com/MaySudo/Misceo

llm-gatewayproxymodel-routingcost-optimizationanthropic-compatibleagentsself-hosted

Also in LLM Gateways & Proxies

diegosouzapw/OmniRoute

OmniRoute is a free AI gateway that unifies access to over 170 AI providers, offering token compression, auto-fallback, and aggregating free tiers to provide billions of free tokens monthly.

Mintplex-Labs/anything-llm

AnythingLLM is an all-in-one local-first AI application for chatting with documents, managing AI agents, and integrating with various LLMs and vector databases, offering dynamic model routing and m...

BerriAI/litellm

LiteLLM is an open-source AI Gateway and Python SDK providing a unified interface to over 100 LLM providers, with features like cost tracking, guardrails, load balancing, and observability.

QuantumNous/new-api

new-api is a unified LLM gateway and AI asset management system that enables aggregation, distribution, and cross-conversion of various LLMs into OpenAI, Claude, or Gemini compatible formats, offer...

Kong/kong

Kong Gateway is a cloud-native API and AI gateway offering high performance, extensibility via plugins, and advanced AI traffic capabilities including multi-LLM support, semantic security, and cach...

decolua/9router

9Router is an AI router and token saver that connects various AI coding tools to over 40 AI providers, optimizing usage with auto-fallback, quota tracking, and token compression.

apache/apisix

Apache APISIX is a dynamic, real-time, high-performance API Gateway that can also function as an AI Gateway, providing AI proxying, load balancing for LLMs, and robust security for AI agents.

lidge-jun/opencodex

OpenCodex is a universal local proxy that enables OpenAI Codex, Claude Code, and Grok Build to utilize any LLM provider, offering advanced model routing, account pooling, and API key management.