Manifest

Backend-as-a-service platform for building data-driven apps with managed infra.

Overview

Manifest is an open-source LLM router designed to make AI agents and applications more reliable. It acts as a single gateway between your agents and multiple LLM providers—whether that's OpenAI, Anthropic, local models, or custom endpoints—allowing you to route requests intelligently, set up automatic fallbacks, and monitor every call.

The core problem Manifest solves is that about 5% of LLM API calls fail in production, and most of those failures are avoidable. With features like Auto-fix (which repairs malformed requests on the fly), model fallbacks (automatic retry on another provider), and detailed observability, Manifest helps teams avoid downtime and understand their inference costs.

You can run Manifest as a managed cloud service or self-host it on your own infrastructure using Docker. Both versions use the same API, so switching between them is just a URL change.

Key features

LLM Gateway with OpenAI and Anthropic-compatible API
Auto-fix: repair broken requests on the fly
Model fallbacks: automatic retry on provider failure
Full request/response logging and observability
Support for subscriptions, API keys, and local models
Cost visualization per agent and provider
Custom routing rules via request headers
Self-hosted and cloud deployment options

Pros

Reduces LLM API failures with auto-fix and fallbacks
No token markup on any plan
Fully open source with MIT license
Supports all major providers and local models
Beautiful dashboard for cost and request tracking
Easy setup: cloud version in under 2 minutes
Free self-hosted option with no usage limits

Cons

Free cloud plan limited to 10,000 requests/month
Auto-fix feature marked as 'coming soon' in some contexts
Requires API keys or subscriptions to providers
Self-hosted version requires Docker setup and infrastructure management

Use this if

You need to route LLM requests across multiple providers, want automatic fallbacks for reliability, need detailed cost tracking, or are building production AI agents that cannot afford downtime.

Skip this if

You only use a single LLM provider and don't need observability, or you prefer a fully managed solution without self-hosting options.

Best for

AI builders connecting multiple LLM providersTeams managing inference consumption and costsPlatforms and providers reducing client errorsProduction AI agents requiring high reliabilityDevelopers needing observability across LLM calls