> ## Documentation Index > Fetch the complete documentation index at: https://docs.murmur.dev/llms.txt > Use this file to discover all available pages before exploring further. # Queue a follow-up > QueueAdd sends a follow-up message to a running agent. Returns NotFound if the agent is not running. ## OpenAPI ````yaml /api-reference/openapi.json post /v1/agent/queue openapi: 3.0.0 info: title: Murmur REST API description: >- PRE-ALPHA: INTERFACE SUBJECT TO CHANGE WITHOUT NOTICE. The murmur REST API lets you spawn agents, track their progress, manage their task lists, and read and write catalog resources from any HTTP client. version: v1 servers: - url: https://api.murmur.dev security: - ApiKey: [] tags: - name: MurmurService paths: /v1/agent/queue: post: tags: - MurmurService summary: Queue a follow-up description: >- QueueAdd sends a follow-up message to a running agent. Returns NotFound if the agent is not running. operationId: MurmurService_QueueAdd requestBody: content: application/json: schema: $ref: '#/components/schemas/v1QueueAddRequest' required: true responses: '200': description: A successful response. content: application/json: schema: $ref: '#/components/schemas/v1QueueAddResponse' default: description: An unexpected error response. content: application/json: schema: $ref: '#/components/schemas/googleRpcStatus' components: schemas: v1QueueAddRequest: type: object properties: agentId: $ref: '#/components/schemas/v1AgentId' description: type: string title: follow-up message tasks: type: array items: $ref: '#/components/schemas/v1Task' title: checklist items to add agent: type: string title: agent persona override followUpId: type: string description: >- Optional client-generated ID for this follow-up. Use it to correlate the follow-up in status and timeline responses. outputSchema: type: string description: >- Optional JSON Schema for this follow-up's structured output. When set, it replaces the agent's active output schema for this turn and remains active for subsequent turns until changed (sticky, like the persona). Empty leaves the current schema unchanged. Validated like SpawnRequest.output_schema. model: type: string description: >- Optional model override for this and subsequent turns (e.g. "claude-opus-4-8"). When set, it replaces the agent's active model and remains active until changed (sticky, like the persona and output_schema). Empty leaves the current model unchanged. Reconciliation with the agent persona is identical to spawn: the explicit model (SpawnRequest.model there, this field here) overrides the persona's AgentConfig.model at the VM. The model must stay on the agent's backend (claude vs codex, inferred exactly as at spawn) — the backend fixes turn dispatch and which provider credentials were sealed onto the VM, so a cross-backend model is rejected with InvalidArgument. reasoningEffort: type: string description: >- Optional reasoning-effort override for this and subsequent turns — sticky, exactly like model. Empty leaves the active effort unchanged. Validated against the agent's backend with spawn's per-backend sets (config.ValidateReasoningEffort): claude accepts low, medium, high, xhigh, max; codex accepts none, low, medium, high, xhigh. An invalid or cross-backend value is rejected with InvalidArgument. serviceTier: type: string description: >- Optional service-tier override for this and subsequent turns — sticky, exactly like model. Empty leaves the active tier unchanged. Codex only (standard, fast, flex — config.ValidateServiceTier, spawn's set); a non-empty tier on a claude agent is rejected with InvalidArgument, mirroring SpawnRequest.service_tier. fastMode: type: boolean description: >- Optional fast-mode override for this and subsequent turns — sticky, exactly like model. Absent leaves the active fast mode unchanged; present sets it (true or false), so fast mode can be switched off mid-session too. Claude's settings.json fastMode, as at spawn (SpawnRequest.fast_mode); not validated per backend, mirroring spawn. preempt: type: boolean description: >- When true, discards any pending queued follow-ups and interrupts the agent's in-flight turn before enqueuing this one — atomically, so the preempting follow-up runs next. The interrupt is best-effort: an idle agent or a turn still starting up is not cut short, but the queue is still cleared and this follow-up still runs as the next turn. deliverAt: type: string format: date-time description: >- When set, defers the message until this time, then it joins the ordinary queue — waking a sleeping agent exactly like a fresh follow-up. Cancel by removing the message with QueueRemove using the returned follow_up_id. Must be strictly in the future and at most 30 days out. Mutually exclusive with preempt. Multiple deferred messages may coexist; re-arm with QueueRemove + QueueAdd. Rejected with FAILED_PRECONDITION when the target agent is on a build without deferral support. v1QueueAddResponse: type: object properties: followUpId: type: string description: >- The message's effective follow-up ID: the request's follow_up_id when supplied (dashboard reconciliation), else a server-minted UUID. Keys QueueRemove and the timeline's QUEUED → DELIVERED/CLEARED transitions. googleRpcStatus: type: object properties: code: type: integer format: int32 description: >- Numeric status code for the failure, matching the HTTP status of the response via the standard mapping (e.g. 3 INVALID_ARGUMENT → 400, 5 NOT_FOUND → 404, 7 PERMISSION_DENIED → 403, 16 UNAUTHENTICATED → 401). message: type: string description: Human-readable description of what caused the failure. details: type: array items: $ref: '#/components/schemas/protobufAny' description: >- Machine-readable error details, when the failure carries them. Usually empty. description: The error envelope returned on every non-2xx response. v1AgentId: type: object properties: tenant: $ref: '#/components/schemas/v1Tenant' ownerProvider: $ref: '#/components/schemas/v1Provider' account: type: string description: >- Owner account: username for identity-provider owners, service-profile name for PROVIDER_SERVICE_PROFILE. agent: type: array items: type: string title: |- Agent path — the hierarchy from root to leaf. Root agent: ["fix-bug"] Child: ["fix-bug", "api"] Grandchild: ["fix-bug", "api", "tests"] workspace: type: string description: >- Workspace this agent belongs to; scopes its repos, environment, and base branches. Optional at spawn: when empty, the server selects one, so a returned AgentId always carries a concrete workspace. description: >- Identifies an agent. The identity is (tenant, workspace, owner, agent path); the workspace scopes the agent's repos, environment, and base branches. v1Task: type: object properties: id: type: string subject: type: string description: type: string status: $ref: '#/components/schemas/v1TaskStatus' blocks: type: array items: type: string title: IDs of tasks this task blocks blockedBy: type: array items: type: string title: IDs of tasks that block this task activation: $ref: '#/components/schemas/v1TaskActivation' persona: type: string description: >- Name of the persona that contributed this task (e.g. "programmer"). Empty for tasks from the spawn request, follow-ups, or task_create. A persona switch removes exactly the persona-tagged tasks and adds the new persona's, leaving all untagged tasks in place. dedupKey: type: string description: >- Stable identity of the external object this task tracks (e.g. a PR review thread). Tasks with the same key are de-duplicated within one delivery batch. Empty for caller-created tasks, which are never de-duplicated. ephemeral: type: boolean description: >- Ephemeral tasks live for exactly one run: created with the follow-up that delivers them and removed when that run completes. Set by the platform on per-thread PR comment and review tasks. description: A checklist item on the agent's task list. protobufAny: type: object properties: '@type': type: string description: >- Identifies the payload type of this detail value; the remaining properties are that type's fields. additionalProperties: {} description: >- A structured detail value. `@type` identifies the payload type; the remaining properties are that type's fields. v1Tenant: type: object properties: provider: $ref: '#/components/schemas/v1Provider' org: type: string title: username or org name description: |- Tenant identifies an organization within a provider. It is the unit of multi-tenant scoping — all resources (agents, pools, VMs) belong to exactly one tenant. v1Provider: type: string enum: - PROVIDER_UNSPECIFIED - PROVIDER_GITHUB_OAUTH - PROVIDER_GITHUB_APP - PROVIDER_GOOGLE_OIDC - PROVIDER_SERVICE_PROFILE default: PROVIDER_UNSPECIFIED description: |- Provider identifies the identity provider type for a tenant namespace. Each provider is a distinct namespace — no cross-provider mixing. - PROVIDER_GITHUB_OAUTH: the namespace of an individual GitHub user — a single login (human or bot, e.g. "alice"), addressed as github_oauth/{login} - PROVIDER_GITHUB_APP: the org/tenant namespace — its account is the org name (github_app/{org}), never a person - PROVIDER_GOOGLE_OIDC: Google Workspace domain - PROVIDER_SERVICE_PROFILE: owner is a CatalogServiceProfile; account is the profile name v1TaskStatus: type: string enum: - TASK_STATUS_UNSPECIFIED - TASK_STATUS_PENDING - TASK_STATUS_IN_PROGRESS - TASK_STATUS_COMPLETED - TASK_STATUS_DELETED default: TASK_STATUS_UNSPECIFIED description: TaskStatus is the lifecycle state of a checklist task. v1TaskActivation: type: object properties: filesModified: $ref: '#/components/schemas/v1FilesModified' description: >- Defines when a task becomes active. At most one condition may be set; unset means always active. v1FilesModified: type: object properties: pattern: type: string title: glob pattern for file paths; empty = any file description: |- FilesModified activates a task when the agent pushes changes. Run-scoped: evaluated against pushes in the current run only. securitySchemes: ApiKey: type: http scheme: bearer description: 'murmur API key: mur_.' ````