AI Agents that don't break
## What is Manifest?
Manifest 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](https://app.manifest.build) and follow the guide.
### Self-hosted
Manifest ships as a [Docker image](https://hub.docker.com/r/manifestdotbuild/manifest). One command:
```bash
bash <(curl -sSL https://raw.githubusercontent.com/mnfst/llm-gateway/main/docker/install.sh)
```
Open [http://localhost:2099](http://localhost:2099) and sign up — the first account you create becomes the admin. Full self-hosting guide: [docker/DOCKER_README.md](docker/DOCKER_README.md).
### Deploy with one click
| Platform | Notes |
| -------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- |
| [Railway](https://railway.com/deploy/wild-wild) | Best path. Template includes Manifest, PostgreSQL, and S3-compatible storage for request recordings. |
| [Render](https://render.com/deploy?repo=https://github.com/mnfst/llm-gateway) | Blueprint includes Manifest, PostgreSQL, and a persistent recording disk. |
| [DigitalOcean](deploy/digitalocean/TUTORIAL.md) | App Platform and PostgreSQL; provide a private Space for recordings. |
| [AWS](deploy/aws/TUTORIAL.md) | CloudFormation provisions ECS, RDS, and a private recording bucket. |
| [GCP](deploy/gcp/TUTORIAL.md) | DeployStack 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](https://manifest.build/docs/deploy/railway), [Render](https://manifest.build/docs/deploy/render), [DigitalOcean](https://manifest.build/docs/deploy/digitalocean), [AWS](https://manifest.build/docs/deploy/aws), [GCP](https://manifest.build/docs/deploy/gcp), [Fly.io](https://manifest.build/docs/deploy/fly), [Coolify](https://manifest.build/docs/deploy/coolify), [Easypanel](https://manifest.build/docs/deploy/easypanel), [Heroku](https://manifest.build/docs/deploy/heroku), [Koyeb](https://manifest.build/docs/deploy/koyeb), and [Apple Containers](deploy/apple-containers/TUTORIAL.md).
> 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.
| Provider | API key / local | Subscription | Model catalog |
| ---------------------------------------------------------------------------------------- | :-------------: | :--------------------------- | --------------------------------------------------------------- |
| [**OpenAI**](https://platform.openai.com/) | ✅ | ✅ ChatGPT Plus / Pro / Team | GPT-5.6 (Sol / Terra / Luna), GPT-5.5, GPT-5.4, Codex, o-series |
| [**Anthropic**](https://www.anthropic.com/) | ✅ | ✅ Claude Max / Pro | Claude Opus 5, Sonnet 5, Fable 5, Haiku 4.5 |
| [**Google**](https://ai.google.dev/) | ✅ | ✅ Sign in with Google | Gemini 3.5 Flash, 3.1 Flash-Lite, Gemini 2.5 |
| [**Google Vertex AI**](https://cloud.google.com/vertex-ai) | ✅ | — | Gemini models via Vertex AI |
| **Gemini Free** | ✅ Managed key | — | Free Gemini models through Manifest's managed gateway |
| [**Meta**](https://www.meta.ai/) | ✅ | — | Muse Spark 1.2 / 1.1 + Contributor route (Meta Model API) |
| [**xAI**](https://x.ai/) | ✅ | ✅ Grok subscription | Grok 4.5, Grok 4.3, Grok Build, Grok 4.20 |
| [**AWS Bedrock**](https://aws.amazon.com/bedrock/) | ✅ | — | Claude, GPT, Kimi, MiniMax, Nemotron, Nova via Bedrock |
| [**Alibaba Cloud / Qwen**](https://www.alibabacloud.com/en/solutions/generative-ai/qwen) | ✅ | ✅ Qwen Token Plan | Qwen 3.7 Max / Plus / Flash, DeepSeek, Kimi, GLM |
| [**DeepSeek**](https://www.deepseek.com/) | ✅ | — | DeepSeek V4 Pro, V4 Flash, V3.2, R1 |
| [**Mistral**](https://mistral.ai/) | ✅ | ✅ Mistral Vibe | Mistral Large, Medium 3.5, Devstral, Codestral |
| [**Moonshot** (Kimi)](https://kimi.ai/) | ✅ | ✅ Kimi Coding Plan | Kimi K3, K2.7 Code, Kimi for Coding |
| [**MiniMax**](https://www.minimax.io/) | ✅ | ✅ MiniMax Coding Plan | MiniMax M3, M2.7, M2.5 |
| [**Xiaomi MiMo**](https://platform.xiaomimimo.com/) | ✅ | ✅ MiMo Token Plan | MiMo V2.5 Pro, V2.5, Flash |
| [**Z.ai**](https://z.ai/) | ✅ | ✅ GLM Coding Plan | GLM 5.2, GLM 5.1, GLM 5 Turbo |
| [**BytePlus**](https://www.byteplus.com/en/activity/codingplan) | — | ✅ ModelArk Coding Plan | Ark Code, Seed Code, GLM, Kimi, DeepSeek, GPT-OSS |
| [**GitHub Copilot**](https://github.com/features/copilot) | — | ✅ Copilot subscription | Claude, GPT, Gemini, Grok via Copilot |
| [**Kiro**](https://kiro.dev/) | — | ✅ Kiro subscription | `kiro/auto`, Claude, DeepSeek, MiniMax, GLM, Qwen |
| [**Command Code**](https://commandcode.ai/studio) | — | ✅ Command Code subscription | Claude, DeepSeek V4, Qwen 3.7, Gemini, Kimi |
| [**ClinePass**](https://app.cline.bot/) | — | ✅ ClinePass subscription | `cline-pass/glm-5.2`, Kimi, DeepSeek, MiMo, MiniMax, Qwen |
| [**NousResearch**](https://portal.nousresearch.com/) | — | ✅ NousResearch subscription | NousResearch Portal model catalog |
| [**OpenCode Go**](https://opencode.ai/) | — | ✅ OpenCode Go | DeepSeek V4, Qwen 3.7, GLM, Kimi, MiMo |
| [**Ollama / Ollama Cloud**](https://ollama.com/) | 🖥️ Local | ✅ Ollama Cloud | Local tags: Llama, Qwen, Gemma. Cloud: DeepSeek V4, GLM, Kimi |
| [**LM Studio**](https://lmstudio.ai/) | 🖥️ Local | — | Local GGUF models, port `1234` |
| [**llama.cpp**](https://github.com/ggml-org/llama.cpp) | 🖥️ Local | — | Local GGUF models, port `8080` |
| [**OpenRouter**](https://openrouter.ai/) | ✅ | — | 300+ models across labs |
| [**OpenCode Zen**](https://opencode.ai/) | ✅ | — | Curated Claude, GPT, DeepSeek, MiMo, Nemotron |
| [**Kilo**](https://kilo.ai/) | ✅ | — | Kilo Gateway catalog |
| [**Cerebras**](https://www.cerebras.ai/) | ✅ | — | GPT-OSS, GLM, Gemma on Cerebras inference |
| [**Fireworks AI**](https://fireworks.ai/) | ✅ | — | DeepSeek V4, Kimi K2.7, Qwen 3.7, Nemotron |
| [**Groq**](https://groq.com/) | ✅ | — | Llama 4, Qwen 3.6, GPT-OSS, Gemma |
| [**Hugging Face**](https://huggingface.co/docs/inference-providers/) | ✅ | — | Open models through Hugging Face Inference Providers |
| [**NVIDIA NIM**](https://build.nvidia.com/) | ✅ | — | Nemotron 3, GLM, Kimi, MiniMax, Qwen |
| [**Pioneer**](https://pioneer.ai/) | ✅ | — | OpenAI-compatible and fine-tuned Pioneer models |
| **Custom** | ✅ | — | Any `/v1/chat/completions` or `/v1/messages` endpoint |
## Quick links
- [Docs](https://manifest.build/docs)
- [n8n community node](integrations/n8n-nodes-manifest) — `n8n-nodes-manifest` on npm
- [Discord](https://discord.com/invite/FepAked3W7)
- [Discussions](https://github.com/mnfst/llm-gateway/discussions)
- [Contributing](CONTRIBUTING.md)
- [GitHub](https://github.com/mnfst/llm-gateway)
## License
[MIT](LICENSE)
