Security
How we keep your prompts safe
mowa only ever touches your prompts — never the code around them — and everything we store is encrypted at rest. Here is exactly what that means.
Report a vulnerability
Found something? Email product@blueskytechnologiescom.com — we acknowledge within two business days. See our security policy for scope.
Prompt-only by design
When mowa scans a repository, it extracts only the prompt text — even when a prompt is embedded inside application logic, mixed in with the surrounding code. We do not read, store, or transmit your other source files. The thing we keep is the prompt, and nothing around it.
We never overwrite your code
Edits are written back as GitHub pull requests that touch only the prompt span. Before committing, a guard verifies that all of your surrounding code is still intact — if anything around the prompt would be lost, the pull request is aborted rather than risk rewriting your file. A prompt change can never silently clobber the rest of your code.
Encrypted at rest and in transit
Prompt content, full version history, integration secrets, and single-use tokens are encrypted with AES-256-GCM before they touch the database. All traffic is served over TLS. Access tokens are random 256-bit secrets stored only as bcrypt hashes, and they are scoped — a write-scoped token can open a pull request for review, never mutate a live prompt directly.
Change control and attribution
Every change to a prompt is versioned and attributed — who changed it, when, from where (editor, GitHub, or an agent) — and flows through review before it goes live. Authentication is Google OAuth or email magic links only; we never store passwords.
Self-hosting is coming
We are building self-hosted mowa, so you can run the whole system inside your own infrastructure and control it end to end — your keys, your database, your network. Until then, we keep your prompts locked down: encrypted, scoped, and prompt-only. Very safe.