One approval queue for every AI action
Agents draft replies, cadence steps and proposals. You approve, edit or reject each one, and the audit log names who did what.
Want to see it first? Book a demo
- Every agent ships in review
- External sends need approval
- Every tool call is logged
The short answer
- SalesCrew puts every AI action that leaves the building into one approval queue. An agent writes the reply, the cadence step or the proposal, attaches its evidence, and stops.
- A person decides. Each item shows the draft, the evidence behind it, a confidence value and its cost, with three actions: approve, edit then approve, or reject with a reason.
- The same queue catches AI clients working over MCP, such as Claude, ChatGPT or Cursor, so an outside agent follows the rules your own agents do.
- Every tool call is written to the audit log, naming whether a person, a token or an agent made it.
Why one queue beats a setting per tool
Most teams switch AI sending off and on tool by tool: one setting in the sequencer, another in the inbox, a third in whatever writes proposals. You end up checking three places, and a new tool arrives with its own default.
SalesCrew keeps one policy and one queue. The policy table decides which classes of action run on their own and which wait. The queue holds everything that waits, grouped by agent, so a morning review takes one pass.
Set it up in five steps
The defaults are already safe; these steps are how you check and tune them.
- 1
Open the Agents page
Every agent is listed with its mode: off, draft, review or auto. Each one ships in review, so nothing sends until you approve it.
- 2
Check the policy table
Under Agents, Guardrails, each action class has a default. External sends are review, spend is human-only, and low-risk work such as tagging and scoring runs in auto.
- 3
Set the guardrails
Choose the per-cadence cap, the per-mailbox and per-domain daily caps and the business-hours window. A contact never gets two touches less than three days apart.
- 4
Work the queue
Items arrive grouped by agent, with the thread, the score or the knowledge-base sources the draft used. Approve it, edit it and approve, or reject it with a reason.
- 5
Promote what proves safe
After a clean week, move a class such as deal creation from positive replies to auto. A draft below its confidence threshold still falls back to review.
What you configure
The default policy is seeded on every instance and edited under Agents, then Guardrails.
| Setting | What it controls | Default |
|---|---|---|
| Agent mode | Per agent: off, draft (writes and stops), review (queues for a person), or auto (runs and logs) | Every agent ships in review |
| Low-risk work | Research, enrichment, tagging, scoring, summaries, internal tasks | Auto |
| Reply handling | Classifying a reply, pausing a cadence on reply, suppressing on unsubscribe or bounce | Auto |
| External sends | Any email, LinkedIn message, SMS or marketplace proposal | Review |
| Record changes | Stage changes, merges and deletes; deals from positive replies; booking a meeting | Review, with the last two free to move to auto after a clean week |
| Spend | Credits, connects, ad budget and anything outside the CRM | Human only, in every mode |
| Guardrails | Days between touches, per-cadence cap, mailbox and domain caps, send window, confidence threshold | Three days between touches; the rest set per instance |
What lands in the queue
The Inbox agent sweeps your inbox every 15 minutes and queues its reply drafts. A marketplace proposal you start with Propose via approval waits there too, with the bid and the cover letter shown in full. Anything an MCP client tries to send arrives the same way.
You can work the queue from Claude or ChatGPT as well as from the app. The approval tools list, approve and reject items with the same effect as the buttons. One kill switch on the Agents page stops every agent at once, and turning agents off never blocks a person working by hand.
For an agency, the audit log answers the question clients eventually ask: what did the AI do on my account, and who checked it? You show them the record.
Questions
- How do I set up human approval for AI-written sales emails?
- In SalesCrew it is already set. Every agent ships in review mode, and any external send defaults to review in the policy table. Open the Agents page to confirm each agent's mode, then check the policy under Agents, Guardrails. From then on, every AI-written email waits in the approval queue for a person.
- Which CRM lets AI agents draft emails but requires human approval before sending?
- SalesCrew does this by default. Agents draft replies, cadence steps and marketplace proposals, and each draft waits in one approval queue until a person approves, edits or rejects it. The same rule applies to an AI client such as Claude or ChatGPT working through the MCP server.
- What happens if nobody approves a queued item?
- It waits. An unattended item stays in the queue and nothing sends on a timer. When a class of work proves safe, you can move it to auto on purpose, and the queue gets shorter.
- Can I run an agent in auto for some actions and review for others?
- Yes, and that is the usual setup. Modes are set per agent, and the policy table sets a default per action class underneath. Research, scoring, summaries and reply classification default to auto. External sends default to review, and spend stays human-only in every mode.
- What does a rejection do?
- The action does not happen, and your reason feeds the agent's next run. A specific reason changes the next draft, so one line on what was wrong is worth writing. Editing before you approve works the same way: the version that goes out is yours.
Run AI with a person on every send
The approval queue and audit log come with every plan. On Core at $299 a month, the Agents pack adds in-product agents for $199.
Want to see it first? Book a demo