kontext

<img src="assets/kontext-banner-cli.png" alt="Kontext banner" width="100%" /> <div align="center"> <p> <a href="https://kontext.security">Website</a> | <a href="https://docs.kontext.security/getting-started/welcome">Documentation</a> | <a href="https://app.kontext.security">Dashboard</a> | <a href="https://discord.gg/gw9UpFUhyY">Discord</a> </p> <p> <a href="LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/license-MIT-152822?labelColor=0d1714"></a> <a href="https://github.com/kontext-security/kontext/releases"><img alt="Latest release" src="https://img.shields.io/github/v/release/kontext-security/kontext?color=152822&labelColor=0d1714"></a> <img alt="Built with Go" src="https://img.shields.io/badge/Go-1.25-152822?labelColor=0d1714"> </p> </div> Runtime governance for AI agents, with local policy decisions, pre-action enforcement, and an authorization ledger. Kontext runs alongside AI agents on developer machines and in cloud environments. It receives tool-use events through local hooks, evaluates policy before consequential actions execute, and records decisions and outcomes in an authorization ledger. The decision path stays local; managed deployments can export redacted records to the Kontext dashboard. [The agent support matrix](docs/coverage.md) is authoritative for each agent and event surface. Blocking is off by default and available only at supported synchronous pre-action hooks. ## Quickstart ```bash brew install kontext-security/tap/kontext kontext setup ``` Create an install token in the Kontext dashboard when setup asks for one. Setup stores the token in the login keychain, installs hooks for supported agents, and starts a background daemon. ```bash kontext doctor ``` Use `doctor` to check self-serve hook status, daemon health, and the managed export backlog. It exits non-zero when a configured installation is unhealthy. For a self-serve stale daemon, `kontext doctor --fix` performs the verified restart; other findings include their manual remediation. Re-run `kontext setup` to rotate the token. Run `kontext setup --uninstall` to remove the self-serve installation. Self-serve setup currently supports macOS. ## How it works ```text agent asks to use a tool | v Kontext receives the action through a hook | v local policy allows, denies, or records a would-decision | v the decision is written to the local authorization ledger ``` The decision path is local. A managed deployment adds configuration and record export; it does not require a hosted service to answer every tool call. ## Core features Kontext balances security and utility for AI agents: low-risk actions keep moving, and unsafe actions can be blocked before they execute. - **Observe agent actions.** Record supported tool calls, policy decisions, and outcomes in a local authorization ledger. - **Apply policy before actions run.** Use deterministic rules for boundaries such as destructive commands, sensitive files, production systems, data exports, and credential access. - **Roll out safely.** Observe mode shows what policy would deny without interrupting work. Enforce mode returns a real denial for matching rules. - **Keep evidence.** Store redacted records locally and, for managed deployments, send them to the Kontext dashboard for review. - **Run where the agent runs.** Use Kontext on a developer machine, in a cloud sandbox, or in another managed agent environment with a supported hook. ## Managed deployments For enterprise identity, audit retention, organization controls, deployment planning, custom usage volume, and onboarding for security and platform teams, contact [michel@kontext.security](mailto:michel@kontext.security) or [book here](https://calendar.superhuman.com/book/11W5Y8b5JsB8dOzQbd/YECs9). ## Agent support matrix [See the agent support matrix](docs/coverage.md) for the exact events, enforcement points, installation scope, and known gaps for each agent. It is the source of truth for what “supported” means; an integration is not treated as fully covered merely because Kontext can receive an event from it. ## Data handling Kontext stores tool activity and decision evidence, not model reasoning or full conversation history. Sensitive values are redacted before local storage and managed export. ## Development ```bash go build -o bin/kontext ./cmd/kontext go test ./... go test -race ./... go vet ./... ``` ## Community - Read [SUPPORT.md](SUPPORT.md) for support channels. - Read [CONTRIBUTING.md](CONTRIBUTING.md) before opening a contribution. - Kontext is released under the [MIT License](LICENSE).