llm-gateway

Manifest LLM Gateway is an open-source LLM gateway for AI agents and apps.

Automatically checked · GitHub checked 2026-11-15

llm-gateway

Overview

Manifest LLM Gateway

AI Agents that don't break

manifest-gh


Deploy on Render Deploy on Railway Deploy on AWS Deploy on GCP

GitHub stars   Docker pulls   Docker image size   CI status   Codecov   license   Discord

mnfst%2Fllm-gateway | Trendshift

What is Manifest LLM Gateway?

Manifest LLM Gateway is an open-source LLM gateway for AI agents and apps. Connect your API keys, subscriptions, and local models to one OpenAI-compatible endpoint, and each query goes to the right model. No single-provider lock-in.

  • 🔀 Custom Routing: API keys, Subscriptions, Local models, Custom providers
  • 💾 Full Body Logs for Success and Error Messages
  • 📊 Track every single dollar, setup notifications and limits
  • 🚑 Fallback on different models when queries fail, Self-heals your bad requests

Quick start

Cloud version

Go to app.manifest.build. Manifest Cloud is $19/month; the self-hosted gateway below is free.

Self-hosted

Manifest ships as a Docker image. One command:

bash <(curl -sSL https://raw.githubusercontent.com/mnfst/llm-gateway/main/docker/install.sh)

Open http://localhost:2099 and sign up — the first account you create becomes the admin. Full self-hosting guide: docker/DOCKER_README.md.

Deploy with one click

PlatformNotes
RailwayBest path. Template includes Manifest, PostgreSQL, and S3-compatible storage for request recordings.
RenderBlueprint includes Manifest, PostgreSQL, and a persistent recording disk.
DigitalOceanApp Platform and PostgreSQL; provide a private Space for recordings.
AWSCloudFormation provisions ECS, RDS, and a private recording bucket.
GCPDeployStack provisions Cloud Run, Cloud SQL, and Cloud Storage.

Every deployment path now uses durable request-recording storage. Railway, AWS, GCP, and Fly.io provision it natively; Render, Coolify, Easypanel, Docker, and Apple Containers mount persistent storage. DigitalOcean, Heroku, and Koyeb collect external S3-compatible settings during setup. Volume-backed templates are single-instance; use S3-compatible storage before scaling horizontally.

Full deployment guides: Railway, Render, DigitalOcean, AWS, GCP, Fly.io, Coolify, Easypanel, Heroku, Koyeb, and Apple Containers.

The old npm-based self-hosting path is no longer supported. Use the Docker image or one of the deployment guides above.

Providers

Manifest connects to 300+ models through 35 built-in provider connections plus any custom OpenAI/Anthropic-compatible endpoint. Bring your own API key, reuse one of 18 subscription flows, or run models locally. Everything is routed through the same OpenAI-compatible endpoint — send "model": "auto" and Manifest picks the model.

Provider catalogs are discovered dynamically when credentials are connected. The examples below are representative, not exhaustive.

ProviderAPI key / localSubscriptionModel catalog
OpenAI✅✅ ChatGPT Plus / Pro / TeamGPT-5.6 (Sol / Terra / Luna), GPT-5.5, GPT-5.4, Codex, o-series
Anthropic✅✅ Claude Max / ProClaude Opus 5, Sonnet 5, Fable 5, Haiku 4.5
Google✅✅ Sign in with GoogleGemini 3.5 Flash, 3.1 Flash-Lite, Gemini 2.5
Google Vertex AI✅—Gemini models via Vertex AI
Gemini Free✅ Managed key—Free Gemini models through Manifest's managed gateway
Meta✅—Muse Spark 1.2 / 1.1 + Contributor route (Meta Model API)
xAI✅✅ Grok subscriptionGrok 4.5, Grok 4.3, Grok Build, Grok 4.20
AWS Bedrock✅—Claude, GPT, Kimi, MiniMax, Nemotron, Nova via Bedrock
Alibaba Cloud / Qwen✅✅ Qwen Token PlanQwen 3.7 Max / Plus / Flash, DeepSeek, Kimi, GLM
DeepSeek✅—DeepSeek V4 Pro, V4 Flash, V3.2, R1
Mistral✅✅ Mistral VibeMistral Large, Medium 3.5, Devstral, Codestral
Moonshot (Kimi)✅✅ Kimi Coding PlanKimi K3, K2.7 Code, Kimi for Coding
MiniMax✅✅ MiniMax Coding PlanMiniMax M3, M2.7, M2.5
Xiaomi MiMo✅✅ MiMo Token PlanMiMo V2.5 Pro, V2.5, Flash
Z.ai✅✅ GLM Coding PlanGLM 5.2, GLM 5.1, GLM 5 Turbo
BytePlus—✅ ModelArk Coding PlanArk Code, Seed Code, GLM, Kimi, DeepSeek, GPT-OSS
GitHub Copilot—✅ Copilot subscriptionClaude, GPT, Gemini, Grok via Copilot
Kiro—✅ Kiro subscriptionkiro/auto, Claude, DeepSeek, MiniMax, GLM, Qwen
Command Code—✅ Command Code subscriptionClaude, DeepSeek V4, Qwen 3.7, Gemini, Kimi
ClinePass—✅ ClinePass subscriptioncline-pass/glm-5.2, Kimi, DeepSeek, MiMo, MiniMax, Qwen
NousResearch—✅ NousResearch subscriptionNousResearch Portal model catalog
OpenCode Go—✅ OpenCode GoDeepSeek V4, Qwen 3.7, GLM, Kimi, MiMo
Ollama / Ollama Cloud🖥️ Local✅ Ollama CloudLocal tags: Llama, Qwen, Gemma. Cloud: DeepSeek V4, GLM, Kimi
LM Studio🖥️ Local—Local GGUF models, port 1234
llama.cpp🖥️ Local—Local GGUF models, port 8080
OpenRouter✅—300+ models across labs
OpenCode Zen✅—Curated Claude, GPT, DeepSeek, MiMo, Nemotron
Kilo✅—Kilo Gateway catalog
Cerebras✅—GPT-OSS, GLM, Gemma on Cerebras inference
Fireworks AI✅—DeepSeek V4, Kimi K2.7, Qwen 3.7, Nemotron
Groq✅—Llama 4, Qwen 3.6, GPT-OSS, Gemma
Hugging Face✅—Open models through Hugging Face Inference Providers
NVIDIA NIM✅—Nemotron 3, GLM, Kimi, MiniMax, Qwen
Pioneer✅—OpenAI-compatible and fine-tuned Pioneer models
Custom✅—Any /v1/chat/completions or /v1/messages endpoint

Quick links

License

MIT