Loom is a managed service. You authenticate to https://api.loom.getmetacognition.com and start ingesting and recalling — there’s nothing to install or run server-side.

​
1. Get your API key

Sign in at loom.getmetacognition.com, open Account → API keys, and create a key. See Authentication for details. 
export LOOM_API_KEY="lk_live_..."

​
2. Ingest a paper

curl https://api.loom.getmetacognition.com/v1/papers \
  -H "Authorization: Bearer $LOOM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"doi":"10.1/astro","title":"Astrocytic calcium signalling","text":"astrocyte calcium gliotransmission ...","oa":true}'

​
3. Recall from memory

curl https://api.loom.getmetacognition.com/v1/recall \
  -H "Authorization: Bearer $LOOM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question":"gliotransmission calcium","budget_tokens":8000}'
Every recalled span carries provenance — the source DOI and section — so your agent always knows where a fact came from.

​
4. Ingest a document

Loom parses PDFs, Excel, Word, CSV, JSON, Markdown, and text — upload the file and Loom turns it into memory. 
curl https://api.loom.getmetacognition.com/v1/ingest \
  -H "Authorization: Bearer $LOOM_API_KEY" \
  -F "file=@paper.pdf" \
  -F "title=My paper"
The response is a full report: characters and tokens extracted, chunks embedded, and claims found.

​
5. Use the Python client

The Python SDK is a thin client over the same API — configure it once and call methods instead of crafting requests: 
pip install loom-sdk

from loom_sdk import Loom

mem = Loom(
    org_id="lab",
    user_id="alice",
    base_url="https://api.loom.getmetacognition.com",
    api_key="lk_live_...",
)

mem.ingest_document("paper.pdf", open("paper.pdf", "rb").read(), title="My paper")
spans = mem.recall("astrocyte calcium")

​
Next steps

Authentication

Keys, tokens, scoping, and errors.

How recall works

Dense + graph + temporal fusion.

Python SDK

Every method on the client.

API reference

All 46 endpoints.