Contextia

Catch secrets before they reach an AI chat. Local and on-device: nothing leaves your machine.

Contextia is a local, privacy-first guard that flags secrets in the chat composer before they ever leave your machine. Developers paste code, logs and configs into AI chats every day, and those often contain live API keys, tokens, private keys and connection strings. Contextia catches them at the moment of typing or pasting, entirely on your device. What it does: • Detects over 80 kinds of credential, covering cloud provider keys, source control and package registry tokens, payment and messaging service keys, AI provider keys, PEM private keys, .env secrets, database connection strings and JWTs. • Optionally flags personal data as well: card numbers validated by checksum, IBANs and national identifiers. Off unless you turn them on. • Explains why each match was flagged, so it is never a mystery. • Underlines the offending text and offers Redact, Allow once, Allow all, or Allow this pattern always. • Lets you add your own values and regular expressions to always redact. • Four modes: Warn, Auto-redact before sending, Block until resolved, or Off. • Keeps a local detections log and counters on your device. Never the secret value itself. Privacy by construction: • Nothing leaves your machine. No accounts, no servers, no telemetry, no third parties. • Zero network requests by default. You may optionally mirror catch counts, never the secret, to a dashboard on your own machine over loopback. It is off unless you enable it and refuses any address that is not loopback. • Runs only on the AI chat sites listed in the permissions, and requests no other access. • Open source under the MIT licence. Audit every line: https://github.com/sbr0nch/contextia Contextia is a safety net, not a guarantee: rule-based detection can miss things, so treat it as a guardrail, not proof that a paste is clean.