Automate the work. Keep the perimeter.
A visual builder for privacy-preserving AI automations. Chain triggers, models and actions, with tokenization on every edge, human-in-the-loop where it matters, and a full audit log of every run.
Tokens through every edge.
Every connection between two steps carries Vault tokens, not raw sensitive data. A model in the middle of the chain never sees a cleartext name, and neither does the step after it.
Confidence gates the robot.
Set a threshold per step. Low-confidence outputs route to a human review queue. Approvals and rejections are audited like any other step, no shadow decisions.
Every run, every step, logged.
Each workflow run is an immutable event stream, trigger, tokenizations, model calls, approvals, dispatches. Replay a run for an investigation, a regulator, or a post-mortem.
Start from a workflow someone already trusts.
24 pre-built workflows for regulated industries, claims triage, KYC red-flag review, contract renewal tracking, support ticket summarization. Each shipped with its privacy-schema, model allow-list, and approval gates wired in. Fork, tune, run.
The model drafts. The human decides.
Set a confidence threshold on any step. If the model's output falls below it, the workflow pauses and the item lands in a reviewer's queue, with the prompt, the tokenized context, and the model's alternatives. Approve, request more context, or reject. Every decision is audited.
What fires. What runs.
Every connector becomes a trigger and an action. Chain them in any order, the privacy layer applies uniformly.
An audit your regulator will recognize.
Every workflow run is a structured OpenTelemetry event stream. Export a time-bounded slice for an investigation, a regulator, or an internal drill, with tokens in place of names, every step of the way.
Your runbook, automated. Your data, where it belongs.
Pick a template, wire your connectors, set your thresholds. Null handles the tokenization, the audit, the handoffs. You handle the work that actually matters.