Manifest: LLM Router for Agents

URL: https://manifest.build/docs/introduction
Type: Open-source developer tool / LLM inference proxy

What It Is

Manifest is an open-source LLM router that sits as a proxy layer between your application and AI providers. It routes each query to the cheapest model that can handle it, claiming up to 70% savings on inference costs.

How Routing Works

The routing engine scores prompts across 23 distinct dimensions to select the optimal provider. Factors include task complexity, specificity requirements, and custom HTTP headers passed by the caller. The system picks the cheapest model that scores above the threshold for the given request.

Provider Support

Four categories:

  1. API key providers — OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, Qwen, Moonshot, MiniMax, Z.ai, OpenRouter
  2. Custom/compatible endpoints — vLLM, TGI, LocalAI (OpenAI- or Anthropic-compatible)
  3. Local models — GGUF models via Ollama, LM Studio, llama.cpp
  4. Subscription-based — existing plans like ChatGPT Plus and Claude Max

API Compatibility

Drop-in compatible with both OpenAI and Anthropic API specs. Functions as a proxy — no SDK changes required.

Cost Controls

Configurable spending limits per agent and per time period (token count or dollar cost). Enforcement options: email alert or hard HTTP 429 block. Prevents runaway inference spend.

Reliability

Built-in fallback chains: if the primary model fails or is unavailable, Manifest automatically routes to the next configured alternative.

Error Handling

Structured error code system (M001–M500) covering authentication, provider config, rate limiting, validation, and internal errors — each with specific remediation guidance.

Deployment

  • Cloud (app.manifest.build) — managed, recommended for quick start
  • Self-hosted — Docker + PostgreSQL, full local control

Key Design Decisions

  • Proxy architecture means zero changes to existing agent/app code
  • Multi-dimensional scoring (23 dimensions) rather than simple cost-rank routing
  • Supports local/subscription models alongside API providers — not just cloud-to-cloud routing
  • Open-source foundation