Feature · Command
Talk
Say what needs doing. Watch your AI employees sort it, build it and ship it.
How it works
Say a request
Open Talk. Select the microphone key to talk hands-free. Or hold the key, or hold Space, while you talk. Release to send. You can also type in the field and press Enter. On a phone, tap the key to talk, or type and tap Send.
Your words show as captions while you talk. The ticket fills in with each word: the title, the kind, the priority, the area, the tags, the scores and the rank. The rules in your browser read each word. The server checks the words when you pause.
When you stop talking, Talk sends the request. A request needs three words or more. The ticket gets a key such as SIG-12. Talk writes known names with their capitals, for example PDF, CSV, Safari, iOS and the first names of your AI employees.
Browsers without speech recognition show the field only. Type your request there. The rest works the same.
Watch triage decide
The triage employee takes the ticket and reads it. Then the decision shows with its reasons, its confidence and its method. The owner shows next to the triage employee, and the light moves from one to the other.
Start the work
Accepted work is Ready to start, unless your organization starts work at once. Select Start, or say "start it". On a phone, Start is over the talk field. When today's model budget is used up, Talk tells you before you start: the work stops at its first model call. Select Raise the budget to change it.
While the work runs, the owner shows with a line that tells what they do now. The progress line stays at Planned until the owner writes the plan. When the owner asks a person a question first, the ticket shows Waits on a person and the question, with a link to answer it. The plan shows as a list. When the owner marks steps as done, the list becomes a checklist with a count. Saved results and checks show when they arrive. The live log shows the newest steps of the owner. Select the count of earlier steps to see all of them on the journey.
When the work ships, Talk shows the total time from your words to shipped. Select Open the result, Receipt or Journey to see more.
Voice commands
Say a command on its own, or at the end of a request after a pause or the word "and":
- "Create it" sends the request now.
- "Start it" starts held work. In a request, the work starts when triage accepts it.
- "Make it urgent", "make it high" or "make it low" sets the priority.
- "That's a bug" sets the kind. Other kinds are feature, idea, docs, incident and research.
- "Assign it to Theo" sets the owner. Use the first name of an AI employee.
- "Not now" puts accepted work in the Backlog.
- "Reject it because …" rejects the ticket with your reason.
- "New ticket" clears the stage for the next request.
A change before triage decides asks for a yes. Say "yes" or "no", or press Enter or Esc.
Several tickets
The live tickets show on the left, or under the top bar on a phone. Select a ticket to bring it into focus. A new request takes the stage, and the earlier ticket stays in the list.
Sounds and spoken replies
Talk plays short, quiet sounds for real events. Select the speaker key to turn them off. Select Speak replies to hear the replies. Spoken replies are off by default. Each spoken word also shows as a caption.
Presenter and speed
Press P to open the presenter. It says a short engineering story through your real organization. Use the arrow keys for the next or the last line, A for autopilot and B to clear. The address /talk?demo=1 starts the story on autopilot. Add &loop=1 to play it again and again.
The story creates real tickets and starts one docs task. Use it in an organization where that is correct.
Press D for the speed readout. It shows how long each step took in this browser: from a word to the card, from a send to the key, and from Start to shipped.
Other pages
Command, Flow and the board are in the top bar. Select the grid key, or press ⌘K or Ctrl K, for every other page.
From the help article Talk: say the work and watch it ship, checked against the product on .
What you can do
The operations this feature's screens call. Each one is POST /api/<name>, and the badge says who may call it.
ViewergetWorkJourneyOne ticket's journey: every step, owner and handoff from the first signal to shipped. Give a workItemId or a signalId.
Types:
GetWorkJourneyRequestreturnsWorkJourneyDto
ViewerlistSignalsLists signals.
Types:
ListSignalsRequestreturnsListSignalsResult
ViewergetTriageSettingsGets triage settings.
Types:
GetTriageSettingsRequestreturnsTriageSettingsDto
ViewergetOverviewGets overview.
Types:
EmptyRequestreturnsOverviewResponse
ViewerpreviewSignalPreviews the signal.
Types:
PreviewSignalRequestreturnsSignalPreviewDto
MembersubmitSignalSubmits a signal.
Types:
SubmitSignalRequestreturnsSignalDto
MemberstartSignalWorkStarts signal work.
Types:
StartSignalWorkRequestreturnsSignalDto
MemberdecideSignalDecides the signal.
Types:
DecideSignalRequestreturnsSignalDto
ViewergetWorkItemGets the work item.
Types:
GetWorkItemRequestreturnsGetWorkItemResponse
ViewergetWorkItemFastPathGets the work item fast path.
Types:
GetWorkItemFastPathRequestreturnsGetWorkItemFastPathResponse
ViewerlistThreadsLists threads.
Types:
ListThreadsRequestreturnsPageResponse<ThreadDto>
ViewergetSignalGets the signal.
Types:
GetSignalRequestreturnsGetSignalResult
AdminNo API tokensupdateTriageSettingsUpdates triage settings.
Types:
UpdateTriageSettingsRequestreturnsTriageSettingsDto
ViewerlistEventsLists events.
Types:
ListEventsRequestreturnsPageResponse<AuditEvent>
ViewerlistEscalationsLists escalations.
Types:
ListEscalationsRequestreturnsPageResponse<EscalationDto>
ViewerlistApprovalsLists approvals.
Types:
ListApprovalsRequestreturnsListResponse<ApprovalDto>
ViewergetPreferencesGets preferences.
Types:
GetPreferencesRequestreturnsPreferencesView
ViewerNo API tokensupdateMyPreferencesUpdates my preferences.
Types:
UpdateMyPreferencesRequestreturnsPreferencesView
ViewerlistDeliveryProofsLists delivery proofs.
Types:
ListDeliveryProofsRequestreturnsListDeliveryProofsResponse
MembergiveFeedbackGives feedback.
Types:
GiveFeedbackRequestreturnsFeedbackDto
ViewergetDocumentRevisionGets the document revision.
Types:
GetDocumentRevisionRequestreturnsDocumentRevisionDto
MemberholdCampaignStop and wait: hold the item and everything under it. Schedules, timers and delegation start no held work.
Types:
CampaignHoldRequestreturnsCampaignDto
MemberreleaseCampaignLift a hold and wake the open items under it.
Types:
CampaignRefreturnsCampaignDto
MembersteerCampaignSend one instruction to every open item in the campaign, for example "keep going but open fewer pull requests".
Types:
CampaignSteerRequestreturnsCampaignSteerResult
AdminsetCampaignSettingsSet the campaign limits (concurrency, total items, spend), delivery mode and pull request limits.
Types:
CampaignSettingsRequestreturnsCampaignDto
ViewergetCampaignReportWhat the campaign did: items by role, runs, pull requests merged and declined, lines changed, cost and follow-ups.
Types:
CampaignRefreturnsCampaignReportDto
MemberlistReviewNotesLists review notes.
Types:
{ workItemId?: string; threadId?: string; reviewerWorkerId?: string; includeSilent?: boolean; limit?: number }returns{ notes: ReviewNoteDto[] }
MemberresolveReviewNoteResolves the review note.
Types:
{ noteId: string; verdicts: ReviewVerdictInput[] }returnsReviewNoteDto
ViewergetWorkItemRoutingGets work item routing.
Types:
GetWorkItemRoutingRequestreturnsWorkItemRoutingDto
MembersetWorkItemRoutingOverrideSets the work item routing override.
Types:
SetWorkItemRoutingOverrideRequestreturnsWorkItemRoutingDto
MembersetStepRoutingOverrideSets the step routing override.
Types:
SetStepRoutingOverrideRequestreturnsWorkItemRoutingDto
ViewergetRunModeGets the run mode.
Types:
GetRunModeRequestreturnsRunModeDto
MembersetWorkItemRunModeSets the work item run mode.
Types:
SetWorkItemRunModeRequestreturnsRunModeDto
MemberupdateWorkItemUpdates the work item.
Types:
UpdateWorkItemRequestreturnsWorkItemDetail
MembertransitionWorkItemMoves the work item.
Types:
TransitionWorkItemRequestreturnsWorkItemDetail
ViewerexplainActionExplains the action.
Types:
ExplainActionRequestreturnsExplainActionResponse
ViewerexplainDecisionExplains the decision.
Types:
ExplainDecisionRequestreturnsExplainDecisionResponse
ViewergetContextManifestGets the context manifest.
Types:
GetContextManifestRequestreturnsGetContextManifestResponse
ViewergetWorkItemModelCallsGets work item model calls.
Types:
GetWorkItemModelCallsRequestreturnsPageResponse<WorkItemModelCallDto>
ViewerlistWorkItemEvidenceLists work item evidence.
Types:
ListWorkItemEvidenceRequestreturnsListResponse<WorkEvidenceDto>
ViewergetProjectFolderGets the project folder.
Types:
GetProjectFolderRequestreturnsProjectFolderView
ViewergetThreadGets the thread.
Types:
GetThreadRequestreturnsGetThreadResponse
ViewerlistWorkItemsLists work items.
Types:
ListWorkItemsRequestreturnsPageResponse<WorkItemSummary>
ViewergetWorkItemTraceGets work item trace.
Types:
GetWorkItemTraceRequestreturnsGetWorkItemTraceResponse
ViewergetWorkItemActionsGets work item actions.
Types:
GetWorkItemActionsRequestreturnsListResponse<ActionDto>
ViewergetWorkItemRepositoryReadinessGets work item repository readiness.
Types:
{ workItemId: string }returnsimport('./repositoryReadiness.js').WorkItemRepositoryReadinessResponse
ViewergetWorkItemSpecThe spec attached to a work item, and the spec blocks this item implements.
Types:
{ workItemId: string }returns{ spec: SpecSummary | null; cites: Array<{ specId: string; specKey: string; blockId: string; ref: string; text: string …
MembercreateSpecWith a work item and no blocks it starts from the item: objective, requirements, plan approach, risks, open questions and recorded decisions. One spec per work item.
Types:
{ title?: string; workItemId?: string; blocks?: SpecBlockInput[]; draftWithAgent?: boolean }returns{ spec: SpecDetail; agentAsked: boolean; reason: string | null }
ViewerlistWorkItemDecisionsLists work item decisions.
Types:
ListWorkItemDecisionsRequestreturnsListWorkItemDecisionsResponse
Screens
What it records
The feature writes these events to the audit log, the permanent record of what happened.
- Policy changed
- Budget reserved
- Checkpoint saved
- Context retrieved
- Escalation raised
- Escalation resolved
- Integration changed
- Memory recalled
- Model called
- Policy evaluated
- Project decision recorded
- Project decision retired