Open a private security advisory on GitHub (https://github.com/diegosouzapw/OmniGlyph/security/advisories/new) or contact the maintainer directly (diegosouza.pw@outlook.com). Do not open a public issue for an exploitable vulnerability.

OmniGlyph is a local proxy between your client (e.g. Claude Code) and the LLM APIs. By design it sees all your session content and your credentials in transit. The corresponding security decisions:

• Binds to loopback by default (127.0.0.1): the dashboard has no authentication and serves captured session context (source text of the images, telemetry). HOST=0.0.0.0 is an explicit opt-in and exposes all of that to the network — only use it on a trusted network.
• Credentials: the proxy forwards the client's auth headers to the upstream and does not persist them. Keys supplied via env (ANTHROPIC_API_KEY etc.) stay in memory.
• Local telemetry: ~/.omniglyph/events.jsonl holds per-request metadata (token counts, body hashes) and, on 4xx errors, compressed body samples — treat the file as sensitive.
• Imaged content is lossy: byte-exact values (secrets, hashes) must never depend on image reads; the pipeline keeps them as text, but the golden rule is: don't put secrets in LLM context.
• Supply chain: pnpm-workspace.yaml enforces a minimumReleaseAge of 3 days for any new package; the core has a single runtime dependency.

Only the latest release line (main / most recent v1.x) receives fixes.