Featured
The default model for most agent workflows in 2026.
Featured
The model to reach for when the job involves images, video, audio, or very long documents.
Featured
When you need token latency measured in milliseconds, not hundreds of milliseconds, Groq is it.
Featured
Still the default for a lot of production agents.
Featured
Useful when an agent needs to swap models without rewiring its caller.
Featured
The default when you want to run a specific open model in production without standing up your own GPU fleet.
View tool→LLM Inference
The Agent Index
The Builder Weekly · Index

The Agent Index

Infrastructure for AI agents. Tools an agent reaches for mid-workflow to do its job. Verified by the builders who use them every day.

  • Cognee

    Open-source memory layer that builds a queryable knowledge graph from an agent's accumulated context, replacing flat vector retrieval with structured semantic recall.

    MemoryPython Library
    Visit →GitHub →

  • Letta

    Stateful agent platform with persistent memory, context management, and a REST API for agent servers.

    MemoryREST API
    Visit →GitHub →

  • Mem0

    Persistent memory layer for AI agents with user-scoped storage, semantic recall, and fact extraction.

    MemoryREST API
    Visit →GitHub →

  • Zep

    Long-term memory for AI agents built on a temporal knowledge graph (Graphiti).

    MemoryPython Library
    Visit →GitHub →
Add to the list

The Agent Index is open source. The canonical list lives on GitHub, and every entry on this page is generated from that repo at build time. Contributions are welcome as pull requests. Read the rules before submitting; the bar is sharp, but the review is fast.

Submit on GitHub →Read the contribution rules
How this list gets made

Tools are added after they show up in production agent workflows or get strong signal from builders with real usage. Inclusion is the editorial judgment. There is no ranking within categories. Every tool listed has earned its place against the same criterion: can an agent complete its full lifecycle through a fully programmatic path?

How we choose tools →