Workspace/Workflow studio
CHECKING ENGINE…LOCAL SANDBOX
AI AUTOMATION, WITH PROOF

Make the busywork
move itself — safely.

Paste a messy request or a whole inbox. RelayProof extracts the details, drafts the follow-ups, and asks for what is missing. Risky actions pause for a human, attacks are blocked, and every run leaves a tamper-evident audit trail.

✳Shield→AI plan→Verify→Human gate→Proof
RUNS SAVED LOCALLY↗
00
No runs yetRedacted audit log
MANUAL WORK PREPARED◷
0 min
EstimatePer-step assumptions
THREATS BLOCKED⊘
0
Injection shield0 gated for humans
AUTOMATION ENGINE✳
LOCAL.
Deterministic rulesGemini optional
01 / INPUT

Request intake

Paste the task that is eating up your time.

NATURAL LANGUAGE
TRY A SCENARIOOne click to load a test request
▣ Emails, phone numbers, cards and secrets are redacted automatically0 / 10000
DRY RUN
◇ No emails sent. No records changed. Every action is simulated.
02 / EXECUTION PLAN

Proof before action

Every step must pass the policy gate.

AWAITING INPUT
⛨ShieldInjection & PIIREADY
⌁PlanIntent & fieldsWAITING
◇Verify6 gatesWAITING
☝Human gateApprove / rejectWAITING
▶SimulateNo side effectsWAITING
▣Policy enforced locallyAction allowlist enabledv1.0.0
03 / HUMAN GATE & OUTPUTS

Decide, then deliver

NO RUN
Outputs appear after a run.
04 / EVIDENCE

Verification report

NO RUN SELECTED
HASH-CHAINED AUDIT TRAIL—
    ⬡Run a workflow to produce verifiable evidence.
    05 / INBOX AUTOPILOT

    Triage a whole inbox in one click

    One request per block — separate with a blank line or a line of ---. Up to 20 at a time.

    BATCH MODE

    Bulk approval is disabled by design: every gated request needs its own human decision.

    0 processed
    ⇶

    No batch yet.

    Load the sample inbox to see six requests triaged at once.
    06 / AUDIT TRAIL

    Run history

    Stored in this browser · redacted 0 records
    07 / GUARDRAILS

    Policy & red team

    HARD BLOCKS — ALWAYS STOP
    RISK SIGNALS — SCORED 0–100 LOW <25 · MEDIUM 25–49 · HIGH ≥50