What Loom stores, and what leaves.

This website

This site serves static files and sets no cookies. There is no analytics script, no tag manager, no session recording, no A/B testing and no advertising pixel. There is no form, and nothing here asks for an email address.

The one outbound request it makes is to GitHub, for the version number on the download page. The host’s own aggregate request counts exist as a function of serving HTTP; nothing in this repository generates or forwards them.

The application

Loom runs on your machine and talks to the model provider you configured. It does not contact any server operated by Aaen Studios, because none exists.

Stored, in ~/.loom: chats, messages and the per-chat workspace index (loom.db); settings, personas, prompts, MCP server definitions, workspaces and the dock layout (config.json); files sent in chats (attachments/); one log per tracked shell command, capped at 5 MB each (logs/); configuration snapshots taken before every change the model makes to Loom’s own configuration, ten kept (backups/); and images produced by the image tool (generated/).

Stored, outside it: provider API keys, in Windows Credential Manager, one entry per provider. They are never written to a configuration file, never logged, and never uploaded.

Set LOOM_HOME to move all of it. There is no sync, no cloud backup and no export that uploads anything.

Command output is logged whether or not you are watching the command, which is worth knowing before running something that might print a secret. The 5 MB cap is per command rather than in total, so eight long-running commands can hold 40 MB between them — in the clear, on your disk.

What leaves your machine

Your conversation, to your provider. Prompts, the files you attach, and whatever a tool asks to read — sent to whichever provider and model you selected, under their terms and their privacy policy. Loom is a client: it does not sit in the middle and cannot see that traffic.

Embeddings, when you index a workspace. File chunks go to your provider’s embedding endpoint so they can be searched semantically. Opt-in per chat, and skipped entirely if you leave it off.

Update checks, to GitHub. Loom fetches a release manifest and verifies its signature. No identifiers, no usage data and no machine fingerprint are sent with it.

That is the whole list. Web search and page fetching go to whichever service is configured — your own Jina key if you have stored one, DuckDuckGo otherwise. Computer use captures screenshots, stored locally and inlined into the request only for the newest capture, so context does not grow with the length of a session.

To send nothing at all: run a local model through Ollama or LM Studio and leave the workspace index off. In that configuration Loom makes no outbound request except the update check, and that can be turned off too.

Telemetry

There is none, and it is not a setting. No crash reporting, no usage statistics, no feature flags, no remote configuration, and no identifier generated to recognise an installation. There is no code path that reports to Aaen Studios. Because the source is public, this is a claim you can check rather than one you have to take on faith.

Your data, and other people’s

Loom is a developer tool with no account system, so there is no personal data held by the publisher to request, correct or delete. Your conversations are on your disk, and deleting ~/.loom is the entire deletion process. Uninstalling deliberately leaves that folder in place, because destroying someone’s history as a side effect of removing a binary would be hostile.

One thing worth knowing: the model can read your other chats, through two read-only tools. Both are refused in Chat mode on purpose, because “answer from the model and the web” must not quietly become “and anything else we have ever discussed”.

Changes

This page describes Loom as it is currently built, not as it might be later. Any change to the claims above is visible in the commit history rather than announced after the fact.