# Manifest

> Manifest is an open-source LLM router that cuts AI agent costs by up to 70%. It is OpenAI-compatible and works with personal AI agents, coding assistants, and app AI SDKs.

Manifest analyzes each query locally in under 2ms and routes it to the right model that can handle it. The routing engine runs entirely on your machine. The cloud version uses a blind proxy that redirects requests to the right provider without reading them. Your data stays yours by architecture, not by policy.

Manifest supports 500+ models across every major provider. Budgets and rate limits, custom providers, model fallbacks, and spend tracking are all built in.

## API contract

- Cloud base URL: `https://app.manifest.build/v1`
- Self-hosted base URL: `http://localhost:2099/v1`
- Auth header: `Authorization: Bearer mnfst_<key>`
- Get a key: https://app.manifest.build
- Request model: `manifest/auto` (the only accepted value; the chosen provider model is returned in the response, e.g. `auto→claude-sonnet-4`. Provider-specific names like `gpt-4o-mini` are not accepted as request models.)
- Endpoints:
  - `POST /v1/chat/completions` (OpenAI-compatible)
  - `POST /v1/messages` (Anthropic-compatible)
- Client SDK: none needed — use the official `openai` or `@anthropic-ai/sdk` packages with `baseURL` set.

## Compatible agents and SDKs

Personal AI agents:
- OpenClaw
- Hermes Agent
- Nanobot
- Craft Agent

Coding assistants:
- Claude Code
- OpenCode

App AI SDKs:
- OpenAI SDK
- Anthropic SDK
- Vercel AI SDK
- LangChain

Anything that speaks the OpenAI-compatible API works with Manifest.

## Docs

- [Documentation](https://manifest.build/docs/introduction/): Full documentation for installing, configuring, and using Manifest
- [Set Up Manifest Cloud](https://manifest.build/blog/setup-manifest-cloud/): Set up Manifest Cloud and start routing agent requests
- [What is an LLM Router?](https://manifest.build/blog/what-is-an-llm-router/): Rule-based vs. AI-powered routing and what it means for autonomous agents

## Product

- [Manifest Cloud](https://app.manifest.build/): Sign up and start routing requests
- [GitHub Repository](https://github.com/mnfst/manifest): Open source codebase, issues, and contributions
- [Pricing](https://manifest.build/pricing/): Local is free forever; Cloud and Enterprise plans
- [Self-hosted](https://manifest.build/self-hosted/): Run the router on your own infrastructure with Docker
- [Free Models](https://manifest.build/free-models/): LLM APIs with permanent free tiers
- [Changelog](https://manifest.build/changelog/): New features, fixes, and improvements, released on GitHub

## Solutions

- [Manifest for AI Builders | Use Your AI Subscriptions](https://manifest.build/solutions/ai-builders/): Connect the AI plans you already pay for to any agent or harness, back them with pay-per-token APIs or local models, and see every dollar spent.
- [Manifest for platforms | Embed LLM routing and Autofix](https://manifest.build/solutions/platforms/): Embed Manifest inside your product. Your users get routing, fallbacks and request Autofix. You keep the surface, the brand and the customer.
- [Manifest for teams | Shared budgets and per-seat visibility](https://manifest.build/solutions/teams/): One gateway for the whole team. Shared budgets, per-seat spend tracking and a single key across every agent your engineers run.

## Supported providers

- [Supported LLM Providers — API Key & Subscription](https://manifest.build/providers/): Every LLM provider routed through Manifest, grouped by how you connect — API key or a subscription like ChatGPT, Claude or GLM — with live model stats.
- [Alibaba Cloud Qwen Models & Routing](https://manifest.build/providers/alibaba/): Alibaba Cloud builds the Qwen models for multilingual and coding work. Route Qwen3 or Qwen2.5 through Manifest with a per agent spending cap and fallback.
- [Anthropic Claude Models & Routing](https://manifest.build/providers/anthropic/): Anthropic makes Claude. Route Sonnet, Opus, or Haiku through Manifest with a spending cap per agent and automatic fallback when Anthropic rate-limits you.
- [Custom Provider Models & Routing](https://manifest.build/providers/custom/): Bring any OpenAI-compatible endpoint — vLLM, Ollama, LiteLLM, or a private gateway — and route it through Manifest with budgets, fallbacks, and spend tracking.
- [DeepSeek Models (V3 & R1) & Routing](https://manifest.build/providers/deepseek/): DeepSeek ships reasoning models at very low prices. Route V3 or R1 through Manifest with a per agent spending cap and automatic fallback.
- [Google Gemini Models & Routing](https://manifest.build/providers/google/): Google makes Gemini. Route 2.5 Pro and 2.5 Flash through Manifest with a spending cap per agent and automatic fallback when Google rate-limits you.
- [Groq Fast LLM Inference & Routing](https://manifest.build/providers/groq/): Groq runs open models like Llama and Qwen on its LPU engine for ultra-low latency. Route Groq through Manifest with a spending cap and fallback.
- [MiniMax Models & Routing](https://manifest.build/providers/minimax/): MiniMax builds long context multimodal models. Route MiniMax M1 or M2 through Manifest with a per agent spending cap and automatic fallback.
- [Mistral Models & Routing](https://manifest.build/providers/mistral/): Mistral, based in Paris, ships open and commercial models. Route Mistral Large or Codestral through Manifest with a cap and automatic fallback.
- [Moonshot Kimi Models & Routing](https://manifest.build/providers/moonshot/): Moonshot builds the Kimi models, known for long context. Route Kimi K2 or K1.5 through Manifest with a spending cap and automatic fallback.
- [NVIDIA NIM Models & Routing](https://manifest.build/providers/nvidia/): NVIDIA NIM serves open models like Llama, Qwen, and DeepSeek. Route them through Manifest with a per-agent spending cap and automatic fallback.
- [OpenAI GPT Models & Routing](https://manifest.build/providers/openai/): OpenAI makes GPT. Route GPT-4o, GPT-4.1, o3, and the rest through Manifest with a spending cap per agent and automatic fallback when OpenAI rate-limits you.
- [OpenRouter Models & Routing](https://manifest.build/providers/openrouter/): OpenRouter puts hundreds of models behind one API. Route any of them through Manifest and let the router pick the right model for each request.
- [xAI Grok Models & Routing](https://manifest.build/providers/xai/): xAI builds Grok, with web and X search baked in. Route Grok 4 or Grok 3 through Manifest with a per agent spending cap and automatic fallback.
- [Z.ai GLM Models & Routing](https://manifest.build/providers/zai/): Z.ai builds the GLM models, a strong open option for coding. Route GLM 4.6 or 4.5 through Manifest with a spending cap and automatic fallback.

## Agents and SDKs on Manifest

- [AI Agents, Coding Assistants & SDKs](https://manifest.build/agents/): Setup guides for OpenClaw, Hermes, Nanobot, Craft Agent, Claude Code, OpenCode, OpenAI SDK, Anthropic SDK, LangChain, and Vercel AI SDK on Manifest.
- [Use the Anthropic SDK with Manifest | Cut Claude API costs](https://manifest.build/agents/anthropic-sdk/): Route every Anthropic SDK call through Manifest with one baseURL change. Cut Claude API costs by up to 70% on apps, agents, and chatbots, in Python or TypeScript.
- [Use Claude Code with Manifest | Cut Claude API costs](https://manifest.build/agents/claude-code/): Route every Claude Code (Claude Agent SDK) request through Manifest. Paste one block into ~/.claude/settings.json and save up to 70% on Claude API tokens.
- [Use Craft Agent with Manifest | Route agent requests](https://manifest.build/agents/craft/): Connect Craft Agent to Manifest with the Manifest provider preset. Use model auto and route requests across your configured AI providers.
- [Use Hermes with Manifest | Route Hermes to Any LLM](https://manifest.build/agents/hermes/): Point Hermes at Manifest and route every request through the most cost-effective model. Setup in 60 seconds with the Hermes config file or the onboard wizard.
- [Use LangChain with Manifest | Route every LangChain call](https://manifest.build/agents/langchain/): Wire LangChain's ChatOpenAI to Manifest in three lines. Route every LangChain agent and chain through the right model.
- [Use Nanobot with Manifest | Route personal agent requests](https://manifest.build/agents/nanobot/): Connect Nanobot to Manifest by adding a custom provider block to ~/.nanobot/config.json. Route every request through Manifest Auto.
- [Use OpenAI SDK with Manifest | Cut OpenAI API costs by 70%](https://manifest.build/agents/openai-sdk/): Route every OpenAI SDK call (including the OpenAI Agents SDK) through Manifest. Set one env var and save up to 70% on token spend without changing your code.
- [Use OpenClaw with Manifest | Cut AI agent costs by up to 70%](https://manifest.build/agents/openclaw/): Connect OpenClaw to Manifest in 60 seconds. Route every OpenClaw request to the right model and save up to 70% on AI tokens. Open source LLM router.
- [Use OpenCode with Manifest | Route coding assistant requests](https://manifest.build/agents/opencode/): Route OpenCode requests through Manifest. Add one opencode.json provider block and use Manifest Auto across your configured AI providers.
- [Use Vercel AI SDK with Manifest | Route Any Model](https://manifest.build/agents/vercel-ai-sdk/): Wire the Vercel AI SDK into Manifest with one createOpenAI call. Route every generateText, streamText, and tool call through the right model.

## LLM error catalog

- [LLM Error Catalog: provider errors, causes & fixes](https://manifest.build/errors/): Browse common LLM provider errors (Gemini, OpenAI, Anthropic, OpenRouter) by provider or error type, with live occurrence data and concrete fixes.
- [Anthropic errors: causes, fixes & live frequency](https://manifest.build/errors/anthropic/): Every Anthropic API error with live occurrence data and concrete fixes. Anthropic 400 — You are out of extra usage; Anthropic 529 — Overloaded.
- [Authentication errors across LLM providers — fixes](https://manifest.build/errors/auth/): Authentication errors across Gemini, OpenAI, Anthropic, OpenRouter and more, with live occurrence data and fixes.
- [Bad request errors across LLM providers — fixes](https://manifest.build/errors/bad-request/): Bad request errors across Gemini, OpenAI, Anthropic, OpenRouter and more, with live occurrence data and fixes.
- [Billing / quota errors across LLM providers — fixes](https://manifest.build/errors/billing/): Billing / quota errors across Gemini, OpenAI, Anthropic, OpenRouter and more, with live occurrence data and fixes.
- [DeepSeek errors: causes, fixes & live frequency](https://manifest.build/errors/deepseek/): Every DeepSeek API error with live occurrence data and concrete fixes. DeepSeek 400 — reasoning_content must be passed.
- [Google Gemini errors: causes, fixes & live frequency](https://manifest.build/errors/gemini/): Every Google Gemini API error with live occurrence data and concrete fixes. Gemini 429 — You exceeded your current quota.
- [Groq errors: causes, fixes & live frequency](https://manifest.build/errors/groq/): Every Groq API error with live occurrence data and concrete fixes. Groq 429 — Rate limit exceeded.
- [Model unavailable errors across LLM providers — fixes](https://manifest.build/errors/model-unavailable/): Model unavailable errors across Gemini, OpenAI, Anthropic, OpenRouter and more, with live occurrence data and fixes.
- [OpenAI errors: causes, fixes & live frequency](https://manifest.build/errors/openai/): Every OpenAI API error with live occurrence data and concrete fixes. OpenAI 429 — Rate limit exceeded; OpenAI 401 — Invalid authentication.
- [OpenRouter errors: causes, fixes & live frequency](https://manifest.build/errors/openrouter/): Every OpenRouter API error with live occurrence data and concrete fixes. OpenRouter 404 — No endpoints found for model.
- [Rate limit errors across LLM providers — fixes](https://manifest.build/errors/rate-limit/): Rate limit errors across Gemini, OpenAI, Anthropic, OpenRouter and more, with live occurrence data and fixes.
- [Server errors across LLM providers — fixes](https://manifest.build/errors/server/): Server errors across Gemini, OpenAI, Anthropic, OpenRouter and more, with live occurrence data and fixes.

## Community

- [GitHub Discussions](https://github.com/mnfst/manifest/discussions): Questions, ideas, and community threads
- [Discord](https://discord.gg/FepAked3W7): Real-time chat and support

## Company

- [About](https://manifest.build/about/): The team behind Manifest
- [Privacy Policy](https://manifest.build/privacy/): How we handle your data
- [Terms of Service](https://manifest.build/terms/): Terms and conditions

## Blog

- [Everyone is building LLM routers, we deprecated ours](https://manifest.build/blog/why-we-deprecated-our-llm-router/): We don't believe in model routing anymore. For most use cases, sticking to a single battle-tested model is the best thing you can do.
- [Request logs are back](https://manifest.build/blog/request-logs/): Request logs are back, opt-in at the agent level. See the full payload on every provider call, including messages, tool usage, and raw JSON.
- [Your failing requests now fix themselves](https://manifest.build/blog/auto-fix-release/): Autofix is live on Manifest Cloud. When a provider rejects a request with a fixable error, Manifest patches it and sends it again. Here's what shipped.
- [The Biggest Reliability Problem in AI Isn't Hallucinations Anymore](https://manifest.build/blog/ai-reliability-not-hallucinations/): In production, the biggest AI problem isn't what the model says. It's whether users get an answer at all. Datadog's 2026 report puts LLM failure at 5%, and 60% of it is rate limits. Reliability is the next competitive advantage.
- [Introducing paid plans for Manifest Cloud](https://manifest.build/blog/introducing-paid-plans/): Manifest Cloud now has a Pro plan at $19/month. The Free plan stays free with all features included. Here is what changes and what stays the same.
- [The reliability stack for LLM agents: tools and methods](https://manifest.build/blog/reliability-stack-llm-agents/): A request can fail before you send it, while it runs, or after it returns. This is a directory of tools and methods grouped by what each one does.
- [This will get you banned from your ChatGPT subscription](https://manifest.build/blog/banned-from-chatgpt-subscriptions/): A ChatGPT subscription is cheap inference, but strictly personal. Sharing, automating, serving other users, or reselling access can get you banned.
- [ChatGPT Plus: Enjoy $200 of Tokens for $20 While It Lasts](https://manifest.build/blog/chatgpt-plus-tokens-third-party-harnesses/): OpenAI lets you spend your $20 ChatGPT Plus subscription inside third-party harnesses like OpenCode and Pi — worth ~$200 in tokens. Here's why it may not last.
- [I Stopped Prompting My Agent. Now I Design the Loop That Prompts It.](https://manifest.build/blog/design-the-loop/): How a Python script, four files, and a strict eval turned a personal AI agent from something I operated into something I supervise.
- [The errors that actually break LLM agents in production](https://manifest.build/blog/llm-agent-errors-production/): Your agent ran clean in the demo. Friday it returns 400s. Here are the six plumbing failures we see most in Manifest's logs.
- [We are deprecating our rule-based routing](https://manifest.build/blog/deprecating-rule-based-routing/): Manifest is deprecating rule-based complexity and specificity routing on September 1, 2026. Why, and how to switch to the default or custom routing tiers.
- [Announcing modelparams.dev, the open-source model parameter database](https://manifest.build/blog/modelparams-open-source-model-parameter-database/): modelparams.dev is a free, open-source database of parameters for popular AI models. Browse the UI, fetch the API, or get type-safe configs in TypeScript.
- [Fable 5 Doubled in Price, Then Vanished in Three Days. Don't Hardwire Your Agents to One Model.](https://manifest.build/blog/fable-5-routing/): Claude Fable 5 launched as Anthropic's most expensive model, then access was suspended in three days. The real lesson: don't depend on a single model — route.
- [Run Claude Code on your ChatGPT Plus subscription](https://manifest.build/blog/run-claude-code-on-chatgpt-plus/): Use Manifest to route Claude Code requests through your existing ChatGPT Plus subscription instead of paying for API keys on top.
- [The Ultimate Guide to AI Subscription Plans](https://manifest.build/blog/subscription-plans-guide/): Connect Claude Max, ChatGPT, Copilot, Gemini, Grok and more to your AI agent through Manifest — use the plans you already pay for instead of per-request fees.
- [10 Ways To Reduce Your LLM API Costs](https://manifest.build/blog/reduce-ai-inference-costs/): Your AI app is live and the inference bill is eating your margins. Here are 10 practical ways to cut LLM costs without hurting output quality.
- [MyTrainer Cuts AI Inference Costs With Manifest](https://manifest.build/blog/mytrainer-manifest-routing/): A practical case study on using Manifest to route coaching chat, deterministic generation, and asynchronous planning workflows in MyTrainer.
- [Anthropic Dropped OpenClaw From Max — What To Do](https://manifest.build/blog/anthropic-kicked-openclaw-users-off-subscription/): Claude Max subscriptions no longer cover OpenClaw usage — every agent message is billed separately. Here are your options and how to route around the change.
- [Set Up Manifest Cloud for your OpenClaw agent](https://manifest.build/blog/setup-manifest-cloud/): Step-by-step guide to set up Manifest Cloud and start routing OpenClaw requests to the right model automatically.
- [Awesome Free LLM APIs](https://manifest.build/blog/awesome-free-llm-apis/): All the LLM APIs with permanent free tiers we could find, updated March 2026. Star the repo or open a PR to help keep it current.
- [What is an LLM Router?](https://manifest.build/blog/what-is-an-llm-router/): An LLM router sends each prompt to a different model based on the task. Here's how rule based and AI routing compare for OpenClaw users.
- [Fix OpenClaw Agents That Won't Stay Autonomous](https://manifest.build/blog/my-openclaw-agent-wouldnt-stay-autonomous/): OpenClaw automations that ran for hours then quietly stopped? Here are the skill files, cron jobs, and memory rules that fix it.
- [Hyper Specialization: Stockfish, Smith & the AI Era](https://manifest.build/blog/hyper-specialization-stockfish-adam-smith/): It's a mistake to think we'll stay relevant by orchestrating AIs. Our salvation may be the opposite: hyper specialization.
- [How to Stop Burning Money on OpenClaw](https://manifest.build/blog/how-to-stop-burning-money-on-openclaw/): After speaking with over a hundred OpenClaw users, cost is the topic that comes up in almost every conversation. Here are the strategies that actually work.
- [How To Shut Up OpenClaw CLI Banner 🦞](https://manifest.build/blog/shut-up-openclaw/): Tired of OpenClaw CLI's banner jokes on every command? Here's the one-line config change that disables them, plus the env var for CI environments.
- [The Claw Market Map, Q1 2026](https://manifest.build/blog/claw-market-map-q1-2026/): The emerging ecosystem built around OpenClaw — key players across managed hosting, LLM routing, security, developer tools, marketplaces and more.
- [Introducing Manifest](https://manifest.build/blog/introducing-manifest/): Introducing Manifest: an open-source LLM router that sends every request to the cheapest capable model and saves AI-native teams up to 70% on token costs.

## Optional

- [Manifest Weekly #09](https://manifest.build/blog/manifest-weekly-09/): Claude Code support via Anthropic endpoint, multiple API keys per provider, and a stack of routing fixes.
- [Manifest Weekly #08](https://manifest.build/blog/manifest-weekly-08/): Manifest Weekly #08: see exactly how much routing saves you in the dashboard, one-command Hermes setup, and a stack of routing and streaming reliability fixes.
- [Manifest Weekly #07](https://manifest.build/blog/manifest-weekly-07/): LM Studio and llama.cpp as built-in providers, custom header routing tiers, simplified setup, and OpenAI Responses API.
- [Manifest Weekly #06](https://manifest.build/blog/manifest-weekly-06/): Manifest Weekly #06: message-level feedback for routing decisions, OpenCode Go subscription provider support, and a pile of routing reliability fixes.
- [Manifest Weekly #05](https://manifest.build/blog/manifest-weekly-05/): Manifest now works with any AI agent, routing by task type, two new subscription providers, and Docker-only self-hosting.
- [Manifest Weekly #04](https://manifest.build/blog/manifest-weekly-04/): Start routing through free models in one click, and Anthropic OAuth finally working with Sonnet and Opus.
- [Manifest Weekly #03](https://manifest.build/blog/manifest-weekly-03/): New cloud onboarding for OpenClaw 3.22-beta, Anthropic OAuth fix for Sonnet/Opus, 100% model pricing coverage, and encrypted API key storage.
- [Manifest Weekly #02](https://manifest.build/blog/manifest-weekly-02/): Copilot & MiniMax subscription providers, smarter model registry, routing scorer improvements, and a pile of bugfixes.
- [Manifest Weekly #01](https://manifest.build/blog/week-12-update/): Anthropic subscription support, automatic model fallback, dynamic model discovery, and a ton of bugfixes.