Demo note: Elasticsearch runs on a 14-day trial; if it has expired, search and indexed data may not load. The UI and demo flow still work.
Workspace
Demo Guide
Demo Guide
Use this page to explain the system clearly to judges before you show the live UI.
What the app does
A human-controlled agent workspace: it turns signals into tasks, drafts actions using SOPs, and keeps an audit trail end-to-end.
1.
Intake (Email + Data)
- Gmail inbound is the primary trigger (push or polling fallback).
- Each thread maps to exactly one parent task: intake.email.
- UI is review/control: inspect, approve, handle exceptions.
2.
Normalize + Index
- Inbox/outbox/audit records are searchable in Elasticsearch.
- This makes retrieval, traceability, and demos repeatable.
- Content stays structured instead of living in messy email threads.
3.
Triage → Task
- Unstructured signals become tasks with type, priority, and risk.
- Each task is routed to the right specialist agent.
- Tasks keep state + events + artifacts for inspection.
4.
SOPs (Playbooks)
- Each task type has an SOP: steps, constraints, and guardrails.
- Agents follow SOPs to stay consistent and safe.
- Humans can update SOPs to improve behavior over time.
5.
Agents Draft Actions
- Specialist agents retrieve context and draft responses/actions.
- Drafts include evidence and a clear rationale.
- Work can create subtasks and handoffs when blocked.
6.
Queue + Worker (Reliability)
- Runs agent work asynchronously via a queue.
- Prevents timeouts and keeps the system resilient.
- Worker writes results back as events/artifacts.
7.
Approvals (Human Control)
- External actions are gated by human approval.
- Humans can edit/reject before anything is sent.
- Designed for trust and safety in production.
8.
Audit + Replay
- Every important step is logged to audit + outbox indices.
- You can answer: what happened, when, by whom, and why.
- This is the foundation for governance and accountability.
Need the wiring? Switch to the Flowcharts tab and click any node for a quick one-liner.