Data residency, redaction, and self-hosted Snapback (coming)
To diagnose a failure, Snapback needs the shape of what happened — the steps, the actions, the errors. It does not need your secrets, and it's built so it never keeps them. Here's how sensitive data is handled today, and what's coming for teams that need the engine inside their own walls.
Redaction happens before the engine sees anything
You don't pre-redact your traces — the service does it for you, server-side, before the diagnosis engine runs:
- Credential fields are masked. Anything keyed as
password,api_key,token,secretand the like is masked out of inputs and outputs. - Secret-shaped patterns are scrubbed. Strings that look like keys —
sk-,AKIA,Bearertokens, JWTs — are scrubbed wherever they appear, not just in fields you labelled. - Responses are sanitised. The verdict is checked so no secret-shaped string is ever returned to you either.
Send the structure of what happened, not your keys. You don't need to strip a trace before sending it — but as good practice, keep raw credentials out of traces and keep your own Snapback token in an environment variable, never in a prompt or a committed file.
Tokens are org-scoped
Every token is scoped to its own organisation — an agent reads and writes only its own org's data. There's no cross-tenant visibility. Your traces, verdicts, and feedback are yours.
The connection is narrow by design
The integration surface is deliberately small: the skill talks only to snapback.sh, runs no shell commands, reads no files, and installs nothing. It's an MCP client calling a diagnosis API — nothing more.
Self-hosted is coming
For teams where data can't leave a boundary at all — strict data-residency rules, regulated environments, air-gapped networks — a self-hosted Snapback is on the way. Run the diagnosis engine inside your own environment, so traces never leave your infrastructure.
If self-hosting is a hard requirement for you, tell us — via request_pattern or the site — and we'll keep you in the loop as it lands. In the meantime, the hosted service's server-side redaction and org-scoped tokens cover the large majority of sensitive-data concerns.
Machine-readable guide: https://snapback.sh/llms.txt · Tool discovery: https://snapback.sh/.well-known/mcp.json · MCP endpoint: https://api.snapback.sh/mcp. Free tools need no token; pay-per-call diagnosis via x402 on Solana or EVM (~$0.0032 USDC).