Loom is a memory API for research agents. You give it papers, notes, and data, and it remembers them — so your agent can recall what it learned, even across separate sessions. You add things to memory with one API call and search them with another. Loom handles the parsing, embedding, and storage for you. Everything you recall comes back with its source attached, so you always know where a fact came from.

Quickstart

Add your first memory and search it in a few minutes.

Authentication

Get an API key and make your first call.

Python SDK

A small Python client for the API.

API Reference

Every endpoint, request, and response.
Loom is a hosted service at https://api.loom.getmetacognition.com. There’s nothing to install or run — get a key and start making calls.

​
What you can do

Add anything

Send PDFs, Excel, Word, CSV, JSON, Markdown, or plain text. Loom turns each one into searchable memory.

Search your memory

Ask a question and get back the most relevant pieces, each with its source.

Link your sessions

Loom connects the sessions that share work and helps you spot useful next questions across them.

​
Good to know

You call Loom over the API or the Python SDK. It doesn’t change how your agent runs — it just gives it a place to remember things.
If a search is slow or unavailable, your agent keeps working. Loom is designed to degrade gently rather than block your pipeline.
Loom keeps track of when facts were true. You can ask what your memory looked like at a point in time, and when two facts disagree, it keeps both.
Each result includes where it came from — the paper and section — so your agent can cite it and you can trust it.