What we do — and don't — know about you.

You don't sign up, log in, or give us an email to chat. Sessions aren't tied to an identity. A peer serving your request doesn't know who you are unless your prompt itself reveals it.

When you chat, your message travels over a TLS-encrypted connection to the mesh entry node and is routed to a peer that runs the requested open-weight model. That peer has to read the prompt to perform inference — this is the honest trade versus a hosted API, and we don't pretend otherwise. Every peer runs the same open-source runtime, so what a peer can and can't do is auditable. For work you don't want to route through anyone else, you can run your own node and keep the entire loop on your hardware.

If no live peer can serve a request, ClosedMesh may fall back to a third-party inference provider to complete it. Those requests are subject to that provider's terms and privacy policy. The source that served any response is exposed in the x-closedmesh-served-by response header (mesh vs fallback) so it's never hidden from you.

• Chat history for the web chat is kept locally in your browser, not on a server account.
• Aggregate metrics — token counts, latency, which models served — are recorded as periodic snapshots to power the public status and metrics pages. These are counts and timings, not the content of your prompts.
• Standard request logs (such as IP address and user agent) may be processed transiently by our hosting to serve traffic and prevent abuse.
• The credits ledger stores contributors' peer IDs and tokens served — not chat content, and not tied to a personal identity.

The mesh checks that a peer actually runs the model it advertises by replaying unpredictable synthetic probes and comparing fingerprints. Only those synthetic probes are replayed across the network — never your prompts.

Running a node contributes your machine's compute to the mesh. Your runtime announces its capabilities (models, backend, available memory) over the mesh, and your node's pseudonymous peer ID appears on the public status page. Other people's prompts pass through your machine to be served; you choose what models to run and can stop at any time.

ClosedMesh is under active development, so this policy will evolve as the product does — we'll update the date above when it changes. Questions or data requests: open an issue on GitHub. See also the security model and terms.