Fact6
Build - Stage 01

From a sentence to structured agent logic.

You write the task the way you'd hand it to a teammate. Fact6 turns that into an explicit plan of actions, conditions, and tool scopes - and shows you every line before anything runs.

agent - live
running
Task
Handle refund requests within our 30-day policy
Connected to
CRM
Stripe
Email
activity.log
Worked example

One task, from sentence to plan.

The plan is not a black box. Every step is named, every tool call is scoped, and every branch is labeled with the condition that triggers it. You can edit the plan in the same language you wrote the task in.

  • Actions are explicit - what the agent will do, in order
  • Conditions are named - when the agent branches or stops
  • Tool scopes are per-action - read vs. write vs. draft-only
  • Escalation targets are declared - who gets pinged, when
plan.agent
v1 - draft
agent handle_refunds {
  when   inbound.ticket.category = "refund"
  guard  order.age_days <= 30
         and refund.amount <= 500

  do
    1. crm.orders.get(order_id)
    2. policy.check(order)
    3. stripe.refunds.create(order, "policy")
    4. email.send(customer, template="refund_ok")
    5. slack.post("#cs-daily", digest)

  escalate
    when refund.amount > 500  ->  slack.dm(@cs_lead)
    when policy.check = fail  ->  human.review
}
PreviousBuild overviewNextTool connections