Questions people ask¶
Where does my data go?¶
To Amazon Web Services, in the US East (N. Virginia) region, and nowhere else: Limma's servers, database and storage all run there. Details: security.
Which models read my data?¶
An open-weights model (Qwen3-30B-A3B-Instruct-2507) that Limma runs on its own GPU server, and Amazon's Titan embedding model on Bedrock for search. No Anthropic or OpenAI model.
Do you train on my data?¶
No. The reader runs fixed, published weights; no model is trained or fine-tuned on your content, and nothing learned from your workspace is used for anyone else.
What are the limits?¶
Documents held, pages and calls a month, workspaces, the largest file: pricing. A page is a PDF page or slide, 3,000 characters read, or 30,000 loaded. The Free plan takes no card: 250 pages a month, for good. Over a limit a call is refused with the reason, never dropped.
How do I delete my data?¶
Delete a document, or a whole workspace, with one call; the reply proves nothing of it is left. Backups keep some of it for longer; the audit archive keeps the record of each call, never its content: deleting.
How do I cancel?¶
In Stripe's customer portal (POST /v1/billing/portal gives you its link). Your plan runs
to the end of the period you paid for; your content is deleted 30 days after it ends.
Which agents work with it?¶
Any MCP client — Claude Code, Cursor, VS Code, Claude Desktop and others — with a few lines of setup, or any program over the HTTP API: get started.
What happens when Limma does not know?¶
It says so. Two documents that disagree come back marked contested; a document it could not read is named, with why; a list that may be incomplete says what is missing.
What can I load?¶
PDFs (scanned ones too), Word, Excel, PowerPoint, CSV, JSON, text, HTML, e-mail and images, from a folder, a bucket, Google Drive, SharePoint or your agent: connect.
Can I run it in my own network?¶
Yes: Limma installs inside your network with your own model, and nothing leaves it. Install it, or write to contact@limma.ai.