Manifest (mnfst/manifest) is a model routing layer for AI agents like Hermes Agent, OpenClaw, and custom pipelines. It intercepts LLM requests, scores task complexity, and routes each request to the cheapest model that can handle it. Simple tasks go to fast, cheap models (Gemini Flash, Claude Haiku, GLM-4.5). Complex tasks go to expensive frontier models only when needed. Docker-based, one-command deploy, MIT license. Supports Anthropic, OpenAI, OpenRouter, Z.ai (GLM-5, GLM-4.7), MiniMax, Mistral, and free OpenRouter models. A companion manifest-ui library provides shadcn/ui components for building ChatGPT-style and MCP applications.
Manifest is a model routing proxy for AI agents. The core insight: most tasks your agent handles don’t require a frontier model. Asking Claude Opus to summarize a short email is paying $5/M tokens for work that Haiku at $0.25/M handles identically. Manifest sits between your agent and your LLM providers, scores each request, and routes it to the cheapest model that meets the task requirements.
Complexity-based routing: Manifest evaluates each request and categorizes it by complexity tier. Simple queries (factual lookups, format transformations, summarization) route to the fastest and cheapest available models. Multi-step reasoning, code generation, and complex analysis route to frontier models only when needed.
Multi-provider support: Anthropic, OpenAI, OpenRouter (including free models), Z.ai (GLM-5/4.7/4.6/4.5), MiniMax, and Mistral are all available as routing targets. The addition of free OpenRouter models means simple tasks can cost nothing.
Zero agent changes: Manifest exposes an OpenAI-compatible endpoint. Any agent or framework already using OpenAI-compatible APIs can point to Manifest instead with a one-line base URL change.
Manifest is most valuable for teams running high-volume agent pipelines where the distribution of task complexity is uneven — most tasks are straightforward, a few require heavy reasoning. Routing correctly gets the 70% cost reduction without sacrificing output quality on the tasks that matter.
Also useful for teams wanting to evaluate multiple providers simultaneously: Manifest’s routing can be configured to A/B test models on equivalent requests and track performance metrics.
The mnfst organization also maintains manifest-ui, a shadcn/ui component library specifically designed for building ChatGPT-style and MCP (Model Context Protocol) applications. Useful for teams building agent UIs on top of their model-routed backend.
Opinionated bundle of agent personas — pre-built role templates for engineering, marketing, ops, and research teams that ship as drop-in markdown files.
OpenAI's flagship AI assistant capable of conversation, analysis, coding, writing, and multimodal reasoning across text and images.
Anthropic's AI assistant known for nuanced reasoning, long-context understanding, safety-focused design, and exceptional writing quality.