Where your code goes, what leaves your machine, and what we store. Written to be accurate rather than reassuring — every claim below reflects how the product actually works.
Agents execute locally on the developer's own machine. File reads, writes, edits, and shell commands all run in your local environment — there is no Prompter Hawk execution server that your code passes through.
Source code leaves your machine only by going to the LLM provider you choose (for example Anthropic, OpenAI, or Google), using credentials you supply. That is inherent to using a cloud LLM and is the same trust boundary as using that provider's own tools.
Three TLS-encrypted channels, none of which carry your source code or file contents:
By default: nothing about your code. We store account, session, and (unless you opt out) telemetry metadata — and that is all.
If you explicitly turn on cloud sync for a mission, we store that mission's event metadata in an encrypted-at-rest Postgres database. Concretely, that means things like file paths (not file contents), the first ~100 characters of shell commands, search patterns, commit messages, and truncated task descriptions. It is an append-only event log, scoped to that one mission. It is not your repository, and it is not your source files.
Access is enforced by mission membership on both push and pull — a request only ever sees missions the requesting user is a member of. On a Team plan, seats are individually licensed; one member cannot read another member's local missions, and cloud-synced missions are visible only to their explicit members.
Your LLM API keys are stored locally on your machine and are passed only to their respective provider. They are never sent to Prompter Hawk. Reading a key from an environment variable requires a one-time explicit consent.
We will answer directly, in writing, and will not overstate our posture. Reach us at security@prompterhawk.dev.