The Missions list
- Lifecycle tabs: All / Active / Paused.
- Stat cards: Total missions, Active, Needs attention, Avg success rate.
- Filters: sort (Last active / Success rate / Avg cost / Name), operational state toggles (Running / Waiting / Paused / Failed), and a department dropdown.
- Toolbar: search, Group (Department / Status / No grouping), time range (All time / 7 / 30 / 90 days), view switcher (List / Compact / Board), Columns (toggle Description, Owner & steps, Trend & metrics), and Export CSV.
- An attention banner highlights missions with a pending approval or a failed last run; an incident banner appears when agent runs failed in the last 24 hours.
- Each row shows status, title, description, creator, step count, a success-rate sparkline, the latest run status, and live progress (“Step 2 of 5”) while a run is in flight.
Creating a mission
Click New mission and choose:The mission page
The mission page opens with a hero — progress ring, status, goal, creator — and these actions:- Run — start a run now (mission must be active and you must have write access).
- Pause / Resume — pause stops new runs.
- Mission Cockpit (✨) — a read-only chat about this mission: “What is the current status?”, “Why did the last run fail?”, “How many credits did the last runs consume?”.
- ⋯ menu — Duplicate Mission (live) and View Capability Manifest; blueprint export/publish are not yet available.
Overview
- Mission health — a 0–100 health score with a headline, plus Current step, Last completed, Next action, Pending approval, and Next scheduled. Buttons: Retry from failed step and Increase budget (add credits and continue a budget-parked run).
- Live KPIs — total runs, success rate, avg duration, credits, avg approval wait, retry rate.
- Mission value — configured business value across runs, with a link to Economics.
- Mission Identity — the Passport (domain, use case, industry) and Genome (structural fingerprint of the flow) at a glance.
Flow
The flow is shown as a horizontal chain: Trigger → steps → Complete, plus an editable step list (drag to reorder, pencil to edit, trash to delete, Add step at the bottom). Step types:
An Agent step is configured with: the agent, a Handoff prompt (context passed into the step), On failure behavior (Retry once then stop / Stop immediately / Continue to next step), an optional Step key (to reference this step’s output in later steps as
{{steps.<key>.output.X}}), and an Output contract — typed fields (string / number / boolean / json, required or not) that the step’s output is validated against on every run.
Click Inspect on a step to open the step inspector: the agent’s mandate, tools & apps, data access, knowledge, memory, permissions, budget, typed inputs/outputs, and per-mission performance — with a link to the agent’s own page.
Runs
- Live — a real-time view of the current run: which step is running or waiting for a human, steps done, a credit budget bar, and controls Stop after current step (graceful) and Cancel now (immediate; side effects of the running step may already exist).
- History — every past run. Select one to see: Summary (goal, result, duration, cost, tokens, agent/mission version), Trace (per-step prompt and answer, attempts, cost), Prompt (the exact assembled system prompt per agent step), Input, Tool calls, Logs (decisions recorded during this run), Evaluation (execution status + business verdict), and Cost (per-step cost table, including credits wasted on failures).
- Logs — the mission-wide decision log with search and level filters (all / info / warn / error).
- Artifacts — files and reports produced by runs, grouped by run.
Approvals
The mission’s own approval inbox. Each pending approval shows the instruction, a risk level, the kind — Mission approval (“may the process advance?”) or Tool approval (“may the agent execute this side effect?”) — and a response deadline. Decisions available: Approve, Approve with conditions (conditions text is required), Reject, Request changes → step (send the run back to an earlier step with a note), and Delegate (hand the decision to another member). Every decision lands in the Discussion thread and the Audit trail timeline. Who may decide: the step’s designated approver, the mission owner, or a lead.Insights
- Analytics — success rate over time, cost over time, and credits by agent, plus quality metrics (avg loop count, tool error rate, human correction rate). Some panels are marked coming soon.
- Economics — business value tracking (gross savings, time saved, total cost, net value, ROI). Savings are configured estimates you provide, not automatically derived. This area is still being rolled out.
What the badges mean. Some mission panels carry a small colored badge: emerald (“ready”) means the panel is freshly wired to your live data; amber (“soon”) means the panel is a preview showing sample data, not your own. Currently on sample data: Analytics (the quality-metrics panels), Cost intelligence, Economics, and the CI / changed-files / artifact details of output contracts. Everything else on the mission page — runs, traces, logs, artifacts, approvals, budgets — is live.
Build
Settings
Who sees a mission
Missions are member-scoped: only the owner, the people added under Members, and Company Admins can see a mission. Department is a label on a mission, not an access boundary. Everyone can open the Missions page and create their first mission.Typical workflows
- Automate a weekly report: Create with AI → review the generated steps in Flow → set a weekly schedule in Settings › Triggers → Run once manually to verify → let it run.
- Put a human in the loop: add a Human approval step before any risky step; the responsible person gets it in their Approvals and Inbox.
- Control spend: set a per-run credit budget; a run that hits the cap parks and asks for a budget increase in Overview.
- Evolve safely: cut a draft in Build › Versions, publish when ready, restore an older version if needed.