limma-ai

The Limma client: the Python SDK for Limma's agent verbs, and the limma command.

Limma gets the right information to an AI agent: every fact at an address, with a receipt that redeems to the exact bytes of the document it came from, and a count that says whether you hold all of them.

Install

pip install limma-ai                  # the SDK (import limma) and the `limma` command
pip install "limma-ai[s3]"            # + reading S3 buckets (also: [azure], [gcs])
pip install "limma-ai[gdrive]"        # + Google Drive      (also: [sharepoint])

The distribution is limma-ai; the package you import and the command are limma.

Keep a folder, a bucket, a Drive folder or a SharePoint library in Limma

export LIMMA_API_KEY=<your key>
limma sync ./contracts --url https://api.limma.ai
limma sync s3://my-bucket/contracts --url https://api.limma.ai --delete --watch 300
limma sync gdrive://<folder-id> --url https://api.limma.ai
limma sync "sharepoint://contoso.sharepoint.com/sites/finance/Shared Documents" --url https://api.limma.ai

Without --workspace the files go to your key's default workspace (created on first use); each pass prints its id. limma sync --help lists every option.

Use it from Python

from limma import LimmaClient

client = LimmaClient(api_key="<your key>", base_url="https://api.limma.ai")
client.ingest("Ingrid Solberg was appointed CFO on 1 March 2024.", filename="minutes.txt")
facts = client.facts("Who is the chief financial officer?")["facts"]
receipt = client.region(facts[0]["source_id"], facts[0]["char_start"], facts[0]["char_end"],
                        page=facts[0].get("page"))

Every method takes workspace_id= to name another workspace your key reaches.

Attach an agent

Agents attach over MCP at https://api.limma.ai/mcp (or your own deployment's /mcp) with the X-API-Key header — no package needed. Documentation: https://docs.limma.ai.