REVCLI · The approval + evidence layer

The approval and evidence layer under every workflow.

The REVCLI platform routes work across providers, pauses every consequential step for a named human's approval, and records replay-grade evidence for the run. It is the deep-dive behind the homepage principle: AI that acts only once you approve it.

Book a diagnosis Watch the approval replay

agent run · sales-pipeline · step 3 of 5

approval pending

revcli agent run sales-pipeline

Loading scope: sales-rep · acme-corp

Step 1/5: CRM evidence read done

Step 2/5: Renewal draft prepared done

Step 3/5: Director approval gate waiting

○ Step 4/5: Send approved proposal

○ Step 5/5: Update CRM and evidence packet

Approval required · Step 3 of 5

Acme Corp Q2 Renewal - $84,000

Reviewer: Sales Director Evidence: diff + timestamp

How it works

Route the run. Gate the risk. Replay the proof.

Every run carries owner, scope, approval state, provider route, and replay evidence.

01 - Route

$ revcli run sales-pipeline --gate high-risk
→ role: sales-rep
→ provider: approved-gateway
→ step 3/5: director approval

Operators launch work from the CLI or web console. REVCLI resolves tenant, team, role, model route, and allowed tools before execution.

02 - Approve

Workflow Sales Pipeline

Step Director review

Reviewer Sales Director

Status Pending approval

Consequential actions stop at the configured gate. Reviewers see actor, system, requested action, diff, deadline, and policy reason.

03 - Replay

14:22:01 workflow.started

14:22:04 tool.crm.read

14:22:09 approval.requested

14:26:37 approval.granted · j.torres

14:26:38 tool.email.send

Each run records tool calls, model route, approvals, outputs, and hashes. Owners replay the run without screenshots or guesswork.

Why now

Agents are crossing from answers into actions.

Buyers need runtime control points: approvals, provider routing, replay evidence, and policy state across existing systems.

Governance gap Oversight lags usage

Autonomous agent adoption is rising faster than operating controls.

Execution layer Control belongs at runtime

Policy should decide which tools, models, and actions each run can use.

Agent workforce Roles need scope

Every AI worker needs owner, team, permissions, cost, and approval rules.

Buying signal Proof beats pilots

The first workflow must show a gated action, the decision, and a replayable record.

Provider-neutral control plane

Govern any AI, any agent, any model route.

REVCLI sits above the systems of record. It assigns work, scopes tools, routes providers, gates consequential steps, and records evidence across the run.

Provider-neutral control plane coordinating governed workflows

Diagnose

Map the workflow, systems touched, risk threshold, reviewer, model route, and client-owned usage cost.

Route

Send each run through approved providers and gateways without tying the buyer to one suite.

Authorize

Check tenant, role, team scope, object access, and approval policy before an action executes.

Execute

Let low-risk steps continue. Stop irreversible or external actions at the configured human gate.

Inspect

Capture actor, tool, command, provider, egress, diff, output, approval, and timestamp.

Replay

Reconstruct what triggered the run, what the agent saw, who approved, and what changed.

Where this runs

The same layer, fitted to your world.

The approval and evidence layer ships as a focused edition for the way you actually work.

Legal

Inside your practice

AI deployed by practice area, configured to your jurisdiction. Each draft cites its source; nothing goes out without your sign-off.

See REVCLI Legal

Small business

One workflow at a time

We install one real workflow in the tools you already use and train your team to run it. You approve before anything goes out.

See REVCLI SMB

Public sector

For public teams

The same approval-gated workflow for government work: every consequential step waits for an authorized human, and the record proves who decided what.

See public sector

Evidence model

Every consequential action has a gate and a replay path.

REVCLI records enough evidence for leadership, ops, security, and the workflow owner to prove what happened after the run.

Owner visibility

Every run captures actor, profile, workflow, command/tool, provider route, approval, egress, output, hash, and replay ID.

Enterprise interface

The web console and CLI are execution surfaces. Personal messaging channels do not trigger governed actions.

Control plane

REVCLI owns catalog, permissions, approvals, policy, and trace correlation. Agents remain the actor inside a workflow, never the authority.

Egress posture

Production autonomy sends outbound HTTP/HTTPS through CrabTrap or equivalent proxy controls before tools touch external systems.

Provider routing

Licensed humans can use Claude Code in their own attended sessions. Shared or autonomous execution uses Anthropic API, Bedrock, Vertex, Foundry, or an approved LLM gateway.

Client-owned usage

The customer pays model, API, cloud, gateway, storage, and deployment costs through their own accounts, with usage tracking and budgets tied to the workflow.

The engine

Built on Claude, an enterprise-grade engine — model-portable by design.

Claude is the engine inside the work. The approval and evidence layer is REVCLI's, so if the engine ever changes, your rules and your record stay exactly the same.

Get started

Find the action your AI should not take alone.

The diagnosis maps one workflow, the action that needs your approval, the reviewer, and the evidence you'll keep before production rollout.

Book a diagnosis Reserve a workshop

Same engine, your rules. The record proves it.