Once configured, a workflow lives in the engine — not in a chat. Agents fire events at each other directly. The LLM is gone. Token bill, zero. Candidate data, never leaves your stack. Multi-day timing is supported by construction.
Every workflow that runs through a chat-based AI agent has three structural failures the moment it leaves a demo. First, every candidate event is an LLM call — token cost scales linearly with funnel volume, and the bill becomes the cost of doing business. Second, every candidate name, every résumé, every voicemail transcript is passed through the LLM provider — a fact that does not survive a GDPR audit or a serious client NDA. Third, the chat session is the runtime; close the tab and the workflow is gone.
The recruiting pipeline cannot live there. It needs to outlive the chat, span days, ignore the LLM after the question of "what do I want to build" is answered. That is what the Workflow Engine is.
After the orchestrator hands off, the engine is the only thing running. Each agent is a small, independent service; signals from one node arrive at slots on another; the engine handles the wiring and the timing.
screened; both the Scheduler and the Dashboard subscribe to it.Agents are not steps in a queue — they are stages in a conveyor. Sourcing keeps surfacing candidates while Matcher ranks the early ones; voice screens fire as soon as the first batch is validated; scheduling starts the moment a screen passes. The same campaign that would take 13 days as a linear handoff finishes in 3 because the engine routes events the instant they happen.
Same engine, very different pipelines. Each one assembled in a chat in under a minute.
10 backend roles at a Series-B fintech. Stack-sensitive, slow to close. Reaches over multiple weeks.
How it runs. Pipeline runs in parallel — sourcing, matching, voice screens, scheduling overlap rather than queue. End-to-end in 3 days instead of two weeks.
150 warehouse staff in a single region. High volume, fast turnaround, mostly mobile candidates.
How it runs. SMS goes out at 9am. Replies trigger short voice screens through the day. Scheduler offers walk-in slots same week. Dashboard updates live.
Two VP-level roles for a portfolio company. Manual outreach, multi-stage interview process, sensitive data.
How it runs. Sourcer surfaces ~40 names. Recruiter approves a shortlist of 8. Scheduler books partner intros. ATS sync writes only the approved candidates — no LLM ever sees the names.
Workflow Master is one product, three concerns. Here are the two that aren't this page.
The conversational layer above the catalog. Describe the pipeline — it picks the agents, sets every property, wires the connections, and starts the campaign. Change the process mid-flight the same way: by talking.
Open →Each HR service is a small, focused agent: sourcing, matching, voice screening, SMS, scheduling, dashboards. Snap any combination together. Swap a piece without touching the rest. Add your own and the orchestrator learns to use it.
Open →On a 30-minute call we'll wire a workflow against a vacancy you have open right now. You close the call. The workflow keeps running.