Coordinate
Runs And Activity
Follow every orchestration run, answer the decisions it needs, ask running agents questions, and manage automations from one Activity page.
A run is a piece of work a coordinator agent plans and worker agents carry out, kept by the runtime host with its plan, tasks, decisions, timeline, and retained worktrees. Activity is where you follow runs, answer what they need from you, ask running agents questions, and manage automations. Orchestration covers the agent side.
Open Activity#
- Choose Activity in the application menu: the Alera menu on macOS, or the Application Menu button beside the app name on Windows and Linux.
- Press
Mod+Shift+A. It works while a terminal has focus.Modis ⌘ on macOS and Ctrl on Windows and Linux. - Click the attention counter in the status bar. Hover it to see which runs, questions, and automations need you, and click a row to jump straight there.
Activity takes over the main area until you choose Return to Workspace or press Esc; your terminals keep running underneath. A rail on the left switches between Runs, Questions, and Automations, each with its own count, and each section remembers where you left it. Selecting a run, a question, or an automation never changes your active workspace; Open Workspace, Open Terminal, and Open Diff are the explicit ways back in.
When a run starts needing you while Activity is not on screen, a Run Needs Attention notification says whether a gate or a plan is waiting; clicking it opens that decision.
Runs#
Runs are grouped by what they need from you:
| Group | A run lands here when |
|---|---|
| Attention | A decision is waiting for you, a task failed, stalled, or is blocked, the coordinator could not start, the run failed, or a worktree, integration, cancellation, or cleanup needs a look |
| Active | It is planning or running and nothing needs you |
| History | It completed, or it was cancelled or failed and everything settled |
Use Search Runs and the project, workspace, and status filters. A run opens on four tabs:
- Overview: decisions waiting for you first, then the objective, the run’s facts (template, isolation, source commit, integration commit, who approved the plan, the coordinator with Open Terminal), and the actions Pause or Resume, Open Workspace, Request Changes, and Cancel Run.
- Tasks: tasks grouped by stage. A task shows its profile, worktree, branch, base commit, dependencies, spec, result, artifacts, validation, and history, with Open Workspace, Open Terminal, Open Diff, and Retry Task. Validation is what the worker reported, not an independent review.
- Timeline: what happened, newest first and grouped by day: dispatches, acceptances, failures and stalls, escalations, decisions, integrations and conflicts, pauses, cleanups, and the messages the run’s agents exchanged. Filter by kind, expand a message or escalation with More, jump to an open decision with Resolve, and open a completed or integrated result with Open Diff.
- Resources (isolated runs): the worktrees and branches the run keeps, and their cleanup. The tab shows how many are still retained.
Start A Run#
- Choose New Run in the Activity header.
- Write the Objective and pick the Workspace. The run starts from the commit shown; uncommitted changes stay where they are.
- Choose the Isolation. Isolated gives every task its own worktree and branch and integrates the results one at a time as squash commits on a local branch. Shared lets workers work in the workspace itself.
- For an isolated run, pick a Template: Freeform (one stage, no gates), Quick Fix, Feature Delivery (with Foundation and Product gates), or one of your own. Manage Recipes… opens Settings → Recipes.
- Choose the Coordinator profile and the worker profiles from your Agent Profiles, and set Concurrent Workers.
- Tick Require Plan Approval to review the plan before any worker starts. Without it the run starts as soon as the coordinator proposes a plan.
- Choose Create Run. The coordinator starts in its own terminal (Open Coordinator Terminal) and prepares a plan.
A run never pushes, opens a pull request, or merges a remote branch. The integrated result of an isolated run stays on a local branch for you to review.
Decisions#
Everything a run needs from you is a card at the top of its Overview:
| Card | What you decide |
|---|---|
| Plan Approval | Approve Plan, Request Changes, or Reject. Review Plan shows the stages, tasks, contracts, and inputs. Rejecting cancels the run. |
| Stage Gate | Once a gated stage is complete and integrated, Approve Gate, Request Changes, or Reject. View Evidence shows the results the approval covers; if they change, you decide again. |
| Stall | A worker showed no activity for 10 minutes. It may still be running, so Alera waits for Kill And Failover, Keep Waiting, or Abort Stage. Show Output shows its last output. |
| Decision | A question a coordinator asked you, with its options or Other… for a free answer. |
Reject and Request Changes need a note. Request Changes sends your note to the coordinator, which proposes a corrected plan; completed work is kept. If a card changed while you were reading it, Alera asks you to refresh before deciding.
Clean Up#
Nothing is deleted automatically. On Resources, once the run has finished, select up to 25 worktrees, opt into Also Delete This Branch per worktree, and choose Preview to see paths, branches, and changed files. Tick Confirm this selection, then Apply Cleanup. A dirty, locked, or busy worktree stops the cleanup until you resolve it; Retry Cleanup and Abandon Cleanup handle one that needs attention.
Questions#
Questions lists what you asked running agents and their replies. Ask Agent picks a running agent, grouped by project and workspace, and shows how the question will reach it:
- Delivers now: the question is pasted into the agent’s terminal right away, or, for a working agent, as soon as its turn ends until its CLI is verified to queue input mid-turn.
- Delivers when its turn ends: the agent gets it when it finishes what it is doing.
- Reads on its next check: a coordinator reads it with its own check.
Choose Deliver to override, and under Advanced set the inbox and Expires If Not Delivered. A thread shows each question and reply, Send Follow-Up, Mark Read, and Cancel Question for one the agent has not received yet. The sidebar’s Ask Agent on an agent row opens the same dialog. Agents answer with an explicit reply; nothing is read from their terminal.
On The Phone#
The mobile companion has the same Activity with Runs, Questions, and Automations. You can follow runs, approve or reject plans and gates, pause, resume, or cancel a run, start a New Run, clean up resources, and ask agents questions. A push for a decision opens the run at that card.
From The CLI#
alera orchestration --json run list --bucket attention
alera orchestration --json run show --run <run-id>
alera orchestration --json run timeline --run <run-id>
alera orchestration --json run new --profile "Opus" --spec "Add audit logging" --template builtIn:feature-delivery --approval
alera orchestration gate resolve --id <gate-id> --decision approve
alera orchestration --json recipe list --workspace <workspace-id>
alera inbox ask --workspace <workspace-id> --agent claude --body "Which tests cover login?" --deliver nowApproving a plan or a gate is your decision: the agent skills tell agents never to resolve gates or approve their own plans.