Most automation platforms ask you to trust them. This one is designed so that you do not have to.
ProcessWeave is installed on hardware you control — your datacentre, or your own cloud account. One deployment serves one customer. There is no shared control plane, no multi-tenant database, and no path by which your process data reaches us. Data residency is a property of where you installed it, not a promise in a contract.
Authentication runs against your identity provider over SSO. Roles, approver pools and task assignment resolve from your own directory and master data. ProcessWeave does not maintain a parallel set of user accounts, and does not become a second place where leavers have to be removed.
Components declare the secrets they need in their manifest and receive them scoped, at execution, inside the worker process. Credentials are never written into a workflow definition, and never passed as a step input — because step inputs are persisted in execution history, and a credential in history is a credential you cannot recall.
Approvals and rejections, every field edit with its value before and after, remarks, manual-entry submissions, and task claims — each with a verified identity attached.
Every extraction and model-derived output, with its inputs, the model used, the output produced, the confidence returned, and the decision the reviewing human made.
Every write to an external system, with the identity of the payload and the outcome, plus retries and escalations.
Append-only here means the service writing the log holds INSERT and nothing else — no UPDATE, no DELETE. Migrations run under a separate credential that the running services do not have. Each service checks its own privileges at startup and refuses to serve if it holds more than it should.
There is no path from a model to your production systems. AI drafts, a human approves, and only then does anything act. Document extraction returns a confidence score for every field, and anything below your threshold goes to a person instead of becoming a guess.
We would rather list these than have you find them in a security review.
Components declare the permissions they claim today; the platform does not yet enforce them. Enforcement is sequenced with the plugin SDK, because it only becomes meaningful once third-party components exist.
The audit trail is append-only by privilege but is not yet hash-chained or written to WORM storage.
Available on request as a direct read; not yet a supported integration.
Retention is currently a deployment decision rather than a product feature.