agentes.chat keeps organizations' model keys and connector credentials, and its agents act with real permissions. We take reports seriously and appreciate coordinated disclosure.
Please do not open a public issue. Instead:
- Use GitHub's private vulnerability reporting: in this repository, go to Security → Report a vulnerability. This is the preferred channel.
- If that's not possible, email security@agentes.chat.
You'll get an acknowledgment within 72 hours. Please include reproduction steps, the version (the
image tag, or version from /api/health) and the deployment type (agentes.chat Cloud or
self-hosted).
This repository (the server, the web app, the runner, and the phone and desktop apps) and agentes.chat Cloud. Most wanted: anything that crosses an organization's boundary, gets past a person's role, an agent's scopes or an approval, or exposes a stored key. A prompt that only makes an agent misuse access it was given is a known limit of agents; report it when our defaults make it easier than they should.
For self-hosted instances, issues in third-party software (Postgres, Docker, your reverse proxy, a model provider) go to those projects, but if our default configuration makes one unsafe, that's ours and we want to know. Fixes ship in new releases: keep instances on the latest one (SELF_HOSTING.md, Upgrades).