The AI never sends without you (unless you want it to)
Choose per action: auto-execute, require approval, or hand off to a human. Every reply, every tool call, every handoff is logged and editable by your team.
The control spectrum
Auto for safe actions. Approval for sensitive ones. Live for the human moments.
Boei runs on its own. Answers questions, looks things up, collects details, sends follow-ups.
Boei drafts the message or the action. You approve, edit, or reject before it happens.
Boei escalates. A human agent joins, sees the full transcript, and finishes the conversation.
Set the level per workflow and per tool. Change it any time. Nothing is permanent.
Boei drafts. You review. Only then does it send.
Turn on approval for the whole workflow, or just for the workflows where tone matters. The AI writes the reply, you get a notification, you approve as-is, edit the text, or reject and give feedback.
- Approve, edit, or reject in three clicks. No typing required.
- Add internal notes to the draft that the customer never sees
- Rejected drafts get your feedback, so the AI drafts differently next time
- Nothing goes out while you're deciding. The customer sees a typing state, not a wrong answer.
Flag any AI action to require your click.
Some actions can't be undone: applying a discount, issuing a refund, sending a quote, changing a booking. Mark those tools as "requires approval" and Boei pauses before executing, showing you exactly what it wants to do with what parameters. You can edit the parameters before approving.
- Set per-tool: safe tools auto-run, sensitive tools wait for you
- Override per workflow if a specific playbook needs stricter gating
- Edit the parameters before you approve (e.g. change €500 to €250)
- Reject with feedback and Boei tries a different approach in the same chat
Boei resumes the conversation with your decision, no re-explaining needed.
The AI knows when to step back.
A customer says "I want to talk to a real person", or the AI runs into a case its instructions don't cover. Boei calls the escalate tool, pings your team, and hands the chat off with a written summary so your agent doesn't have to read the whole thread.
- Customer keywords ("live chat", "real person") trigger escalation directly
- The AI can decide to escalate based on the situation (via the escalate tool)
- Agent receives an AI-written handoff summary in the notification
- If no one is available, a friendly fallback goes out and the chat waits in queue
- Contact: Liam Anderson · [email protected]
- Topic: Garden room quote, 20 m²
- Where we are: Estimate shared (€24,900). Customer asked for a discount, escalated per policy.
- Suggested next step: Offer 5% loyalty discount OR site visit next Tue 10:00.
Route to the right person automatically.
Pick who is on shift, how much they can handle, and how new chats reach them. Boei handles the routing.
Distribute new chats evenly across available agents. Or use random.
Alert the whole team and let the first available agent grab it.
Each agent controls their own on/off status. Off-shift agents are skipped.
Cap how many chats one agent can hold at a time. No one gets buried.
Reach agents where they actually look.
Every agent picks their own channels and hours. Escalations always break through; routine chats respect the schedule so nobody gets pinged at 2am for a low-priority ping.
- Per-agent notification schedule (working hours only, or 24/7)
- Escalations override the schedule (email + Slack always fire)
- Optional central team email for shared inboxes
Invite your team. Grant only what they need.
Add collaborators one by one and tick the permissions that fit their job. Someone who just answers chats doesn't need billing access. Someone in ops doesn't need to touch the AI training.
Internal notes stay internal.
Tag a teammate, leave context, warn the next agent about a fragile customer. The visitor never sees any of it.
Know exactly who decided what, and when.
Every approval, edit, rejection, escalation and takeover is logged with the person's name, timestamp, and (for rejections) their feedback. Useful for coaching new agents, defending a decision, or auditing what the AI actually did on your behalf.
- Every AI-drafted reply is stored with its approval decision
- Every tool call records inputs, outputs, and human overrides
- Handoff summaries live in the thread, not a separate log
- 14:02 · AI drafted reply
- 14:03 · Sarah approved with edit ("changed 24,900 to 24,500")
- 14:07 · AI proposed
apply_discount(500) - 14:08 · Sarah rejected ("too much for a first-time discount, try 5%")
- 14:09 · AI retried with 5%, auto-approved by rule
- 14:11 · Mark took over live chat
"I was suspicious of it for nearly a whole week, sending endless texts and emails to myself in utter disbelief, when it failed to keel over. Set up is straight forward, easy and takes minutes. Seriously, I cannot recommend Boei enough."
Give the AI its work. Keep the last word.
Start with everything on approval. Turn off the guardrails one at a time as trust grows. Full control from day one.