manifest.build
manifest.build is The backend AI can generate, developers can trust!
Last verified:
What is manifest.build?
manifest.build is Manifest is a smart model router for agents and AI applications that redirects each query to the right model, saving up to 70% in AI inference costs. It analyzes the complexity and specificity of each request in under 2 milliseconds and routes it to the cheapest model capable of handling it, rather than using a single expensive model for all tasks. Manifest is built for developers using AI coding tools like Cursor, Copilot, and Windsurf, as well as anyone building AI apps or running personal AI agents who wants to control inference costs.
manifest.build pricing
Pricing model: Freemium
Basic version is free to use with no credit card required. You can connect your own API keys from supported providers (BYOK) and start routing requests at zero Manifest cost. Free API tiers from providers like OpenRouter work out of the box. You pay usage costs directly to the provider based on their pricing. Cloud version is recommended for newcomers with easy setup. Self-hosted version based on Docker is free and gives full control. Subscription providers can be connected to use quotas first before falling back to pay-as-you-go.
manifest.build pros
- Saves up to 70% on AI inference costs
- Smart routing based on query complexity and specificity
- Four routing tiers: simple, standard, complex, reasoning
- Routing scoring happens in under 2 milliseconds
- Up to 5 fallback models per tier for outage recovery
- Real-time dollar spending visualization
- Soft and hard limits with email alerts
- Supports all major LLM providers out-of-the-box
- Drop-in OpenAI API replacement - just change endpoint
- Free tier available - no credit card required
- Connect your own API keys (BYOK) from any provider
- Supports subscription providers to use quotas first
- Local model support via Ollama, LM Studio, llama.cpp
- Model benchmarking to compare quality, price, latency
- Fully open source - cloud and self-hosted versions
- Session momentum remembers last 5 tier assignments
- Custom HTTP header routing for specific control
- Works with AI coding tools: Cursor, Copilot, Windsurf
- Replay past queries to optimize model selection
- No complicated folder structures or config files
manifest.build cons
- Self-hosted version requires Docker and technical expertise
- Local models only available in self-hosted mode
- Cloud version runs on their servers (less privacy control)
- Per-attempt timeout defaults to 180 seconds may be too long
- Hard limit returns HTTP 429 which may break some clients
- Email alerts sent hourly via cron may have delay
- Routing updates managed by Manifest team (cloud version)
- Maximum 5 fallback models per tier limitation
- 424 status does not trigger fallback (prevents loops)
- Requires NodeJS 20.x or superior to run locally
Frequently asked questions about manifest.build
Which LLM providers and models does Manifest support?
Manifest supports all major LLM providers out-of-the-box, including OpenAI, Anthropic, MiniMax, DeepSeek, Mistral, Google, xAI, Qwen (Alibaba), Moonshot (Kimi), Z.ai (Zhipu), and OpenRouter. You can also connect any provider that has an OpenAI-compatible API. The self-hosted version allows you to run local models too using Ollama, Llama.cpp, and LM Studio as first-class providers.
Can I use my pro subscription with Manifest?
Yes, Manifest lets you connect your subscription to make the most of it. As subscriptions often have lower API rate limits, Manifest recommends adding fallback models to complete requests when the subscription model fails. Manifest provides easy connection for popular subscriptions like Anthropic, GitHub Copilot, MiniMax, Ollama Cloud, OpenAI, OpenCode Go, and Z.ai.
Do I need to pay to use Manifest?
No, the basic version is free to use. You can connect your own API keys from supported providers (BYOK) and start routing your requests without paying Manifest anything. You only pay the usage costs directly to your chosen AI provider based on their token pricing.
How does routing work?
Manifest scores each query across 23 dimensions grouped into keyword-based (14), structural (5), and contextual (4) categories. The weighted sum maps to one of four tiers: simple, standard, complex, or reasoning. Scoring happens in under 2ms with zero external calls. Custom routing uses HTTP headers, while default routing sets primary and fallback models. Session momentum remembers the last 5 tier assignments with a 30-minute TTL.
Can I use Manifest with free models?
Yes. Manifest supports free models from OpenRouter and other providers. You can mix free and paid models in your routing configuration. Simple tasks go to free models while complex ones go to paid ones. Several cloud providers offer free API tiers that work out of the box with no credit card required to get started.
What happens if my primary model fails?
You can add up to 5 fallback models per tier. If a provider returns any HTTP 4xx or 5xx error (including 429 rate limit, 500 internal error, 503 service unavailable) or times out, Manifest automatically retries with the next fallback model in the chain. Your agent keeps working and gets a response instead of an error. The response includes X-Manifest-Fallback-From header showing which model failed.
Do I need to adapt my agent setup?
No, Manifest is a drop-in replacement for OpenAI's API. You just need to change the endpoint to the Manifest URL. Manifest provides tutorials for SDKs like Anthropic's SDK, OpenAI's SDK, and the Vercel AI SDK, as well as agents like OpenClaw and Hermes, and coding assistants like Claude Code.
What's the difference between cloud and self-hosted?
The cloud version is hosted on Manifest's servers and is easier to set up and maintain, recommended for newcomers. The self-hosted version runs on your machine or infrastructure using Docker and gives you more control and privacy. Local model providers (Ollama, Llama.cpp, LM Studio) are only available in self-hosted mode. Both versions are fully open source.
How do I set spending limits and alerts?
Navigate to your agent's Limits page in the dashboard and click Create rule. You can create Email Alert rules (sends email when threshold hits), Hard Limit rules (returns HTTP 429 and stops requests), or both. Choose a metric (tokens or cost), a threshold amount, and a period (hour/day/week/month). Rules are evaluated hourly for notifications and on every request for blocks.
What are the four routing tiers?
Simple tier handles greetings, definitions, and short factual questions routed to cheapest models. Standard tier handles general coding help and moderate questions for good quality at low cost. Complex tier handles multi-step tasks, large context over 50k tokens, and code generation using best quality models. Reasoning tier handles formal logic, proofs, math, and multi-constraint problems using only reasoning-capable models.