Workflow Engine

AgentLane runs the agent flows itself, inside the AgentLane API. Partners and clients never see or edit the flows: they configure an agent, AgentLane executes it.

What runs your agents

AgentLane runs the agent flows itself, inside the AgentLane API. Each catalog agent is a fixed, tested sequence of steps (look up the client’s configuration, ask the AI model, send an SMS or an email, check or book a calendar slot). When an event arrives, the API queues a run and a worker executes the steps in order on the same server as the rest of the platform. There is no separate automation product to log into, and no third-party automation service receives the data that flows through it.

Run payloads (customer phone numbers, e-mail addresses and message text) are kept for 30 days and then removed.

SMS messages and bookings are at most once: if a step is retried after a failure, it never sends the same message twice, and a lost outcome raises an alert instead of being repeated. Messages that would go out during a client’s quiet hours wait until the next morning in the client’s local time.

Provisioning a partner

Each partner goes through a tracked provisioning pipeline, visible to Admins at /admin/automation/[partnerId]. See Admin: Provision a partner for the operational walkthrough.

unprovisioned ──Provision──► provisioning ──► ready
                                  │
                                  └──failure──► failed ──Provision (retry)──► provisioning

ready ──Reprovision──► provisioning   (rebuilds the partner's flows)
ready ──Deprovision──► unprovisioned  (permanent — cannot be undone)

Every attempt is broken into individually-timed pipeline steps, each recording success/failure, warnings, and both an internal error and a client-safe error message, so a partner-facing “provisioning failed” retry button never has to leak infrastructure detail, while Admins debugging the same failure see the full internal error and step timing. If a partner’s agent request fails to provision, see Troubleshooting → Provisioning failures.

Provisioning binds each requested agent to its flow and enables it. Credentials are looked up when a step runs and are never copied into the flow.

Separation between partners is enforced by AgentLane’s own authorization checks.

How a request flows end to end

1. Configuration: a Partner requests an agent for a client from /dashboard/agents; once fulfilled, they configure client-specific values (Twilio number, Cal.com link, business hours, services) from that client’s detail page. See Deploy your first agent.
2. Trigger: an external event occurs (an inbound Twilio call, a missed call, a chat message) and hits the AgentLane API via webhook.
3. Execution: the API queues a run for that client’s flow with the event payload, and a worker runs the steps.
4. Logging: the result is written back to AgentLane, which populates the Executions feed and rolls the result up into that client’s usage counters and the agency’s Analytics dashboard.

Partners never author flows directly. The Portal lets them deploy pre-built agents by filling out configuration forms and mapping credentials.