TensorChat is pre-1.0. Security fixes land on the latest release only; there are no backported patch branches yet. No version carries a support commitment, and the software is provided as is, without warranty, as the LICENSE sets out — nothing on this page changes that.

Do not open a public issue for a security vulnerability.

Report it privately through GitHub Security Advisories, or by email to support@tensorspace.ai.

Please include:

• What the issue is and what an attacker gains from it.
• Steps to reproduce, ideally against a fresh cargo run -p tensorchat-server.
• The version or commit you tested.
• Any proof-of-concept you have.

This is a small project with no staffed security team, so no response time is promised. Reports are read and triaged as maintainers are available. Where one is agreed to be a vulnerability, the aim is to fix it before it is described publicly and to coordinate timing with the reporter, who is credited in the advisory and changelog unless they ask not to be. None of this is a commitment or a term of service; it describes intent.

The server, the store, the wire protocol, and the web client — anything in this repository. Particularly:

• Authentication and session handling.
• Authorization: any path where a user reaches a channel, message, or file they are not a member of or entitled to.
• Cross-site scripting, cross-site request forgery, and Content-Security-Policy bypasses.
• Anything reachable through the upload path.
• SQL injection, and denial of service that a single unprivileged account can cause.

• Deployment misconfiguration. Binding TC_BIND=0.0.0.0:8080 directly to the internet without TLS is documented as the operator's responsibility.
• Missing rate limits outside authentication. Only /api/login and /api/register are rate limited today; this is known, and general request throttling belongs at the reverse proxy.
• The known limitations listed in the README — single-node operation, the user directory shipping whole at connect, and so on. These are design boundaries, not vulnerabilities.
• Reports generated solely by an automated scanner with no demonstrated impact.
• Social engineering, physical attacks, and anything requiring a compromised host.

Two things matter most when running TensorChat exposed:

Terminate TLS in front of it. The server speaks plain HTTP. Session tokens are bearer credentials; over an unencrypted connection they are readable in transit.

Raise the auth rate limits behind a proxy. The limiter keys on the client peer address. A reverse proxy that presents every client as the same address will make one user's failed logins throttle everyone — set TC_AUTH_BURST and TC_AUTH_PER_SECOND accordingly, and rate limit at the proxy instead.