Your agent decides what to say. SalesLobe decides whether it has earned the right to send, with an audit trail of every decision. Connect any AI system to the SalesLobe API and every message it writes goes through the same governance layer as a human rep.
Available on Pro and Agency plans.
Anyone can wire an LLM to an inbox. The hard part is everything around it: who approved this message, does it stay within the client's tone, what happens when the model gets it wrong, and can you prove, later, who wrote what. A raw agent has no gate, no trail, and no way to earn trust. That's not an AI problem. It's a governance problem.
A reply comes in
Your workspace receives the reply. SalesLobe fires a signed webhook (lead.replied) to your endpoint, and if Corty analyzed it, also a second event (corty.suggested) with Corty's draft, confidence score, and classification. Both include the reply_id your agent needs.
Your AI decides
Use Corty's draft as context, write its own reply, or skip. Your agent, your logic, your models. SalesLobe doesn't care what your AI is built with.
It submits the send
POST /v1/replies/{reply_id}/send with the message body. One endpoint, one API key, scoped to exactly what your agent needs.
The governance gate decides
No active autonomy grant? The message does not send. It's stored as a pending draft in the review queue, visible in the dashboard and waiting for a human. With earned autonomy, it sends immediately. Either way, the decision is logged.
The first time your agent calls the send endpoint, this is what it gets back:
HTTP 202. Accepted, not sent. Your agent's draft is in the review queue; a human approves, edits, or rejects it in the dashboard. Nothing your AI writes reaches a prospect until it has earned that right. Handle 202 as success with pending, not as an error: it means the governance layer is doing its job.
{
"status": "queued_for_review",
"reply_id": "dd235c7e-…",
"reason": "no_active_autonomy_grant"
} Agent Control Plane · Earned Autonomy
No separate rulebook for AI. Corty, your human reps, and your own agent all climb the same L0 → L3 ladder, earn autonomy the same way, and can be pulled back to zero the same way. No exceptions, no bypass.
Teams are wiring LLMs to inboxes and campaigns faster than they can govern them. A raw agent with send access has no gate and no track record. That's not an AI problem, it's a governance problem.
So every external agent starts at Level 0: all sends queue for human approval. As its drafts prove out, approved unchanged and aligned with your tone rules, the org owner grants it more autonomy, until approved sends go out immediately. Pull it back to zero at any moment.
How Earned Autonomy worksHuman. Agent. Corty. Same ladder.
Positions are earned per actor: any of them can reach L3, and any of them can be pulled back to L0.
Teams already ship custom agents to their inboxes and campaigns. A raw LLM with send access is a compliance risk waiting to happen, not a feature.
Nobody manages trust agent-by-agent or rep-by-rep. Every actor proves itself on the same ladder, so autonomy means the same thing everywhere.
The EU AI Act's transparency obligations start landing in August 2026. Provenance recorded at the moment of sending, not detected after the fact.
Each send attempt, queued or sent, is written to an append-only decision log: which actor (your agent, identified by its API key), whether a grant was active, what action resulted, and when. If your agent edited Corty's suggestion, the edit distance is recorded too. The full chain is visible per reply in the dashboard as a Decision Trail: suggested, submitted, queued, approved, sent, each step with its actor and timestamp.
We don't detect AI-written text after the fact. We record it at the source.
The EU AI Act's transparency obligations begin applying in August 2026, and the direction is clear: organizations deploying AI that interacts with people will increasingly need to show when and how AI was involved. You can't retrofit that onto sent email; AI-written text carries no watermark, and detection tools are unreliable. What holds up is provenance recorded at the source: who drafted, who edited, who approved, who sent. That's what the decision log gives you, per message, from day one.
SalesLobe provides an audit trail, not legal advice; how the AI Act applies to your use case is a question for your counsel.
Alignment
Pass original_suggestion and your agent's final text with the send, and SalesLobe measures the distance between what Corty proposed and what your agent decided. Over time that becomes a signal: an agent whose edits shrink is converging on the workspace's voice, the same alignment evidence a human rep produces, and the same basis for earning autonomy. Two models, one governance layer, one score.
BYOK
Corty's classification and drafting can run on your own Anthropic API key. You pay inference directly, and the key is encrypted and never exposed. Available on Pro plans and above.
Your agent runs on whatever stack you choose, on your own infrastructure and your own model bills. SalesLobe never sees its internals. You use the API, and Neural Credits cover the platform and governance layer.
Pricing
A full pipeline run (classification, Corty's draft, governed send) is 1 Neural Credit. Sends where your agent brings its own text are metered separately; that pricing is being finalized during the partner pilot. Webhook delivery and read requests are not charged.
Create an API key in your workspace
Scope it to only what the agent needs: leads.read, replies.send, webhooks.manage. Don't use a full-access key for an agent.
Register your webhook endpoint
POST /v1/webhooks, then verify the HMAC signature on incoming events.
Handle the events
lead.replied and corty.suggested. Fetch the full reply via GET /v1/replies/:id when you need more than the truncated preview.
Send via the API
POST /v1/replies/:id/send, and treat 202 queued_for_review as success.
Build a track record
Your first sends land in the review queue. Approve them in the dashboard, build a track record, earn autonomy.
Direct Gmail and Outlook send integration, so governed autonomy extends to day-to-day inbox replies. And we're exploring, with our pilot partners, native connectors so agents and assistants can discover and use SalesLobe without writing HTTP: same governance, less plumbing. If you're building in this direction, we want to hear from you.
Apply for the partner pilotYou don't need to apply to start building. The API docs are open, and Level 0 (reviewed sending) works with any Pro API key today. The pilot is for teams ready to earn autonomy.
Connect it this week: first sends reviewed, autonomy earned, every decision on the record.