API Reference

Complete Rover docs for workspace key management, RoverBook Firestore data paths plus private settings callables, public task resources, and browser runtime contracts.

Public task base: https://agent.rtrvr.ai/v1/tasks · Runtime base: https://agent.rtrvr.ai/v2/rover/*

Auth Modes

Firebase ID token for Workspace management APIs (/generateRoverSiteKey, /listRoverSiteKeys, and related config endpoints).
sessionToken (rvrsess_*) for browser runtime calls to /v2/rover/* after session bootstrap.
publicKey (pk_site_*) only for bootstrap exchange on /v2/rover/session/open. Do not send sk_site_* to browser runtime.
task access token for the public task URL returned by /v1/tasks.
Firebase auth + Firestore rules for owner-only RoverBook analytics reads in Workspace, plus Firebase-authenticated callables for private settings such as notification subscriptions.

Public Agent Task API (`/v1/tasks`)

Neutral AI / CLI protocol for Rover-enabled sites. The source-visible marker <script type="application/agent+json"> is optional but recommended for raw HTML discovery. Host-only launches still work when callers post directly to https://agent.rtrvr.ai/v1/tasks with a canonical envelope like { url, goal, capabilityId, args, target, identity, execution, accept, policy }. Compatibility aliases such as { url, prompt } and { url, shortcutId } still work.

POST/v1/tasks

Create a neutral public task for any Rover-enabled site using a target URL plus a canonical goal or shortcut envelope.

Auth: Anonymous public access token minted in response. Initial create call requires no site public key.

Required Request Fields

Field	Type	Description
`url \| targetUrl`	`string`	Target page URL on a Rover-enabled site.

Optional Request Fields

Field	Type	Description
`goal`	`string`	Canonical natural-language goal for the task.
`prompt`	`string`	Compatibility alias for `goal`.
`shortcutId \| shortcut`	`string`	Exact saved shortcut/capability ID. Use for repeatable flows.
`capabilityId`	`string`	Explicit page/site capability identifier when the caller already knows the target skill.
`args`	`Record<string, unknown>`	Structured task arguments for the selected capability.
`target`	`{ pageId?, route?, capabilityId?, entityId?, formId? }`	Optional page/entity/form targeting hints.
`identity`	`{ userPresent?, signature?, signatureInput?, verificationMethod? }`	Structured caller/user-presence and verification hints.
`execution`	`{ mode?: 'auto' \| 'browser' \| 'cloud'; userPresent?: boolean }`	Requested execution path and user-presence hints.
`accept`	`{ modes?: ('text' \| 'markdown' \| 'json' \| 'observation' \| 'artifacts')[] }`	Requested result modes.
`policy`	`{ confirmation?: 'auto' \| 'required' \| 'site_policy'; crossHost?: 'allow' \| 'same_host' \| 'site_policy' }`	Execution-policy hints that the site may honor or narrow.
`agent`	`{ key?, name?, vendor?, model?, version?, homepage? }`	Optional self-reported visiting-agent metadata used by Rover and RoverBook attribution.

Success Example

{
  "id": "agt_123",
  "task": "https://agent.rtrvr.ai/v1/tasks/agt_123?access=agt_access_...",
  "workflow": "https://agent.rtrvr.ai/v1/workflows/wf_456?access=wf_access_...",
  "open": "https://www.rtrvr.ai/#rover_receipt=rrc_...",
  "browserLink": "https://www.rtrvr.ai/?rover=get+me+the+latest+blog+post#rover_receipt=rrc_...",
  "status": "pending"
}

Notes

Headers drive advanced behavior: `Accept`, `Prefer: wait=15`, `Prefer: execution=auto|browser|cloud`, and `Idempotency-Key`.
Compatibility aliases like `prompt` and `shortcut` still work, but `goal`, `shortcutId`, `capabilityId`, `args`, `target`, `identity`, `execution`, `accept`, and `policy` are the canonical task shape.
If no explicit `agent` object is provided, Rover can still attribute the caller heuristically from `User-Agent`, `Signature-Agent`, `Signature`, `Signature-Input`, and `X-RTRVR-Client-Id`.
Default execution mode is `auto`: browser attach is preferred. `open` is the clean receipt-based browser handoff, while `browserLink` is an optional readable alias when the visible deep link stays within a conservative URL budget.
Use `Prefer: execution=cloud` for guaranteed browserless execution today. Explicit cloud tasks omit browser handoff URLs.

GET/v1/tasks/{id}

Read the canonical public task resource as JSON, SSE, or NDJSON.

Auth: Task access token via `?access=` or `Authorization: Bearer ...`.

Notes

`Accept: application/json` returns latest state or final result.
`Accept: text/event-stream` streams uniform task events.
`Accept: application/x-ndjson` streams the same events in CLI-friendly NDJSON.

POST/v1/tasks/{id}

Continue a task that is waiting on user input.

Auth: Task access token.

Required Request Fields

Field	Type	Description
`input`	`string`	User or agent continuation answer.

DELETE/v1/tasks/{id}

Cancel an in-flight public task.

Auth: Task access token.

POST/v1/tasks/{id}/handoffs

Create a delegated child task on another Rover-enabled site and keep the same workflow lineage.

Auth: Parent task access token.

Required Request Fields

Field	Type	Description
`url`	`string`	Target Rover-enabled site URL.

Optional Request Fields

Field	Type	Description
`instruction`	`string`	Delegation instruction for the target site.
`goal`	`string`	Canonical goal for the delegated child task.
`prompt`	`string`	Compatibility alias for delegated natural-language work.
`shortcutId`	`string`	Exact shortcut to run on the receiving site.
`capabilityId`	`string`	Explicit child-task capability identifier.
`args`	`Record<string, unknown>`	Structured arguments for the delegated capability.
`target`	`{ pageId?, route?, capabilityId?, entityId?, formId? }`	Optional page/entity/form targeting hints on the receiving site.
`identity`	`{ userPresent?, signature?, signatureInput?, verificationMethod? }`	Structured user-presence and verification hints for the child task.
`execution`	`{ mode?: 'auto' \| 'browser' \| 'cloud'; userPresent?: boolean }`	Requested execution path for the child task.
`accept`	`{ modes?: ('text' \| 'markdown' \| 'json' \| 'observation' \| 'artifacts')[] }`	Requested child-task result modes.
`policy`	`{ confirmation?: 'auto' \| 'required' \| 'site_policy'; crossHost?: 'allow' \| 'same_host' \| 'site_policy' }`	Execution-policy hints for the child task.
`agent`	`{ key?, name?, vendor?, model?, version?, homepage? }`	Optional child-task agent identity override. If omitted, the current visiting-agent attribution is inherited.
`contextSummary`	`string`	Structured summary to carry into the child task.
`expectedOutput`	`string`	Describe what the child should return to the parent workflow.

Notes

Delegated child tasks are still ordinary task resources; they simply inherit the parent's workflow lineage.
Receiving sites must enable `siteConfig.aiAccess.allowDelegatedHandoffs` in Workspace.

GET/v1/workflows/{id}

Read the aggregated cross-site workflow resource as JSON, SSE, or NDJSON.

Auth: Workflow access token via `?access=` or `Authorization: Bearer ...`.

Notes

`Accept: application/json` returns the latest workflow snapshot and final result when available.
`Accept: text/event-stream` or `Accept: application/x-ndjson` streams workflow-level lineage across parent and child tasks.

Workspace Management APIs

Used by Rover Workspace for key lifecycle and site config management. These remain Firebase-authenticated control-plane endpoints.

Cloud site config persists shortcuts, business type, sparse Rover V3 experience overrides, greeting, voice dictation, AI access policy, agent discovery policy, and per-site page-capture overrides. Runtime-owned defaults still control the seed CTA, bottom-center anchor, transparent live-action mode, and default live-card count unless you explicitly override those fields. Boot-only flags such as ui.mascot.soundEnabled and ui.muted still stay outside these workspace endpoints.

POST/generateRoverSiteKey

Create a new Rover site key and persisted policy/profile.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`siteId \| siteName`	`string`	Existing siteId or a siteName for server-generated siteId.
`label`	`string`	Human-readable key label.
`allowedDomains`	`string[]`	Allowed host/domain patterns (`example.com`, `*.example.com`, `=app.example.com`, or URL-shaped entries normalized to host). In `registrable_domain`, plain `example.com` covers the apex host and subdomains.

Optional Request Fields

Field	Type	Description
`ttlDays`	`number`	Expiration days, use 0 for no expiry.
`environment`	`'production' \| 'development' \| 'test'`	Deployment environment label.
`capabilities`	`{ roverEmbed?, externalWebContextScrape?, cloudAgent?, cloudScrape? }`	Capability profile for this key.
`roverPolicy`	`RoverSitePolicy`	Policy persisted and returned in list/rotate APIs, including `siteMode`, `domainScopeMode`, navigation policy, and shortcut limits.

POST/listRoverSiteKeys

List all Rover site keys for the authenticated workspace user.

Auth: Firebase ID token (Bearer)

POST/updateRoverSiteKeyPolicy

Patch a key's allowed domains, active status, capabilities, and Rover policy.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Target key identifier.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion/update.
`allowedDomains`	`string[]`	Replacement domain pattern set (`example.com`, `*.example.com`, `=app.example.com`, or URL-shaped entries). In `registrable_domain`, plain `example.com` covers the apex host and subdomains.
`active`	`boolean`	Enable/disable key.
`capabilities`	`ApiKeyCapabilities`	Capability patch.
`roverPolicy`	`Partial<RoverSitePolicy>`	Policy patch, including `siteMode`, `domainScopeMode`, and navigation policy updates.

POST/rotateRoverSiteKey

Rotate a key (invalidate old key and issue a new pk_site_* value).

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Key being rotated.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion/update.
`label`	`string`	Optional new label.
`capabilities`	`ApiKeyCapabilities`	Capability patch applied to rotated key.
`roverPolicy`	`Partial<RoverSitePolicy>`	Policy patch applied to the rotated key, including `siteMode`, `domainScopeMode`, and navigation policy.

POST/deleteRoverSite

Delete a Rover site, all owned site keys, and Rover-owned config while retaining historical RoverBook analytics.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`siteId`	`string`	Rover site identifier to delete.

Notes

Deletes all owned Rover site keys for the site, plus Rover-owned site config and RoverBook owner settings/questions.
Historical RoverBook analytics, visits, reviews, interviews, notes, posts, and aggregates are retained.
Webflow-managed sites are detached from the live managed Webflow install before Rover-owned resources are removed.

POST/getRoverSiteConfig

Read cloud shortcuts, business type, sparse experience overrides, greeting, voice, AI access, discovery, and page-capture config for a key/site.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Site key ID.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion.

POST/upsertRoverSiteConfig

Write cloud shortcuts, business type, sparse experience overrides, greeting, voice, AI access, discovery, and page-capture config.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Site key ID.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion.
`siteConfig.shortcuts`	`RoverSiteShortcut[]`	Persisted shortcuts/journeys.
`siteConfig.businessType`	`RoverBusinessType`	Tunes runtime placeholder hints and generated quick actions when the site has few explicit shortcuts.
`siteConfig.experience`	`RoverSiteExperience \| null`	Sparse Rover V3 surface overrides only. Runtime defaults still own the seed, centered stage, and focus stream when a field is omitted. Use null to clear overrides.
`siteConfig.greeting`	`{ text?, delay?, duration?, disabled? } \| null`	Greeting config.
`siteConfig.voice`	`{ enabled?: boolean; language?: string; autoStopMs?: number } \| null`	Browser dictation config. autoStopMs is the post-speech silence window in milliseconds. Use null to clear persisted voice settings.
`siteConfig.aiAccess`	`RoverSiteAiAccess \| null`	Public launch/runtime guardrails for ATP task entry, browser deep links, cloud browser use, and delegated handoffs.
`siteConfig.agentDiscovery`	`RoverSiteAgentDiscovery \| null`	Persisted public discovery policy, well-known artifact paths, host-surface binding, and seed/presence-led display settings. Use null to clear.
`siteConfig.pageConfig`	`RoverPageCaptureConfig \| null`	Sparse per-site page-capture overrides. Use null to clear overrides.

POST/updateApiKeyCapabilities

Patch capability flags for user or site keys.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Target key identifier.
`capabilities`	`ApiKeyCapabilities`	Boolean capability patch fields.

RoverBook Workspace Data Paths

RoverBook analytics pages now read Firestore directly under owner-auth rules. Backend callables remain only for private settings that include masked secrets, such as notification subscriptions and interview-question management. These are distinct from public site-tag ingest routes and are not public `siteId` browser APIs.

READFirestore: roverbook_* collections

Owner-facing RoverBook analytics now read Firestore directly under ownership rules instead of calling a backend read facade.

Auth: Firebase auth + Firestore rules (`ownerUid` scoped).

Notes

Workspace reads `roverbook_sites`, `roverbook_visits`, `roverbook_scores`, `roverbook_reviews`, `roverbook_interviews`, `roverbook_questions`, `roverbook_notes`, and `roverbook_posts` directly.
Signed public RoverBook GET routes still exist for runtime/agent surfaces used by `@rover/roverbook`.

POSTgetRoverBookNotificationSettings

Read the private RoverBook setup payload for a site, including interview questions and masked webhook subscriptions.

Auth: Firebase ID token (owner only).

Required Request Fields

Field	Type	Description
`siteId`	`string`	Owned Rover site identifier.

POSTupsertRoverBookNotificationSettings

Write RoverBook interview questions and webhook subscriptions for a site.

Auth: Firebase ID token (owner only).

Required Request Fields

Field	Type	Description
`siteId`	`string`	Owned Rover site identifier.

Optional Request Fields

Field	Type	Description
`enabled`	`boolean`	Enable or disable RoverBook outbound notifications for the site.
`webhooks`	`RoverBookWebhookSubscription[]`	Private per-site webhook subscription list.
`interviewQuestions`	`string[]`	Per-site RoverBook interview prompts.

Notes

Webhook secrets are masked on read and preserved unless explicitly replaced.
These settings are stored privately by owner + site and are not exposed through public embed config.

POSTupsertRoverBookQuestions

Write just the RoverBook interview questions for a site.

Auth: Firebase ID token (owner only).

Required Request Fields

Field	Type	Description
`siteId`	`string`	Owned Rover site identifier.
`interviewQuestions`	`string[]`	Per-site RoverBook interview prompts.

Rover Runtime APIs (`/v2/rover/*`)

Server-authoritative runtime contract for embedded Rover. All run-state changes are keyed bysessionId + runId + epoch + seq.

POST/v2/rover/session/open

Bootstrap or refresh a browser runtime session and return the initial projection.

Auth: Body token. Preferred: bootstrapToken/publicKey (pk_site_*). Optional: sessionToken refresh path.

Required Request Fields

Field	Type	Description
`siteId`	`string`	Workspace site identifier.
`host \| url`	`string`	Current host/url for domain and policy checks.
`bootstrapToken \| publicKey`	`string`	Public Rover key (pk_site_*). Required when no valid sessionToken is supplied.

Optional Request Fields

Field	Type	Description
`sessionId`	`string`	Client session identifier (auto-generated if omitted).
`sessionToken`	`string`	Current rvrsess_* token. If invalid/expired and bootstrap token exists, server falls back to bootstrap flow.

Success Example

{
  "success": true,
  "data": {
    "sessionId": "visitor-123",
    "sessionToken": "rvrsess_...",
    "sessionTokenExpiresAt": 1771484403000,
    "streamToken": "rvrsess_...",
    "epoch": 5,
    "capabilities": { "roverEmbed": true },
    "policy": {
      "domainScopeMode": "registrable_domain",
      "externalNavigationPolicy": "open_new_tab_notice",
      "crossHostPolicy": "same_tab",
      "enableExternalWebContext": true,
      "externalScrapeMode": "on_demand",
      "externalAllowDomains": [],
      "externalDenyDomains": []
    },
	    "projection": { "sessionId": "visitor-123", "epoch": 5, "events": [], "tabs": [] },
	    "siteConfig": {
	      "shortcuts": [],
	      "businessType": "saas",
	      "experience": {
	        "presence": { "ctaText": "Do it with Rover" },
	        "inputs": { "voice": false }
	      },
	      "greeting": { "text": "Welcome" },
	      "voice": { "enabled": true, "language": "en-US", "autoStopMs": 2600 },
	      "aiAccess": { "enabled": true },
	      "agentDiscovery": {
        "enabled": true,
        "preferExecution": "auto",
        "roverSiteUrl": "/.well-known/rover-site.json",
        "agentCardUrl": "/.well-known/agent-card.json",
        "discoverySurface": {
          "mode": "beacon",
          "branding": "site",
          "hostSurface": "auto",
          "actionReveal": "click",
          "beaconLabel": "Use AI"
        }
      },
      "pageConfig": { "disableAutoScroll": true, "adaptiveSettleMaxWaitMs": 480 }
    },
    "sseUrl": "https://agent.rtrvr.ai/v2/rover/stream?..."
  }
}

Typed Conflict/Auth Errors

Example 1

{
  "success": false,
  "error": "SESSION_TOKEN_EXPIRED",
  "data": {
    "code": "SESSION_TOKEN_EXPIRED",
    "message": "Token expired",
    "retryable": true,
    "next_action": "Pass bootstrapToken/publicKey (pk_site_*) to /v2/rover/session/open to re-bootstrap the session."
  }
}

Example 2

{
  "success": false,
  "error": "BOOTSTRAP_REQUIRED",
  "data": {
    "code": "BOOTSTRAP_REQUIRED",
    "message": "bootstrapToken/publicKey (pk_site_*) is required.",
    "retryable": false,
    "next_action": "Provide a valid pk_site_* key in bootstrapToken/publicKey."
  }
}

Notes

Server mints short-lived sessionToken (~10 min).
Browser runtime should not send long-lived bearer keys.

POST/v2/rover/session/open

Refresh an existing rvrsess_* token and SSE URL.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Session to refresh (defaults to token sid).

Notes

Use when token is near expiry or after SSE auth failures.

POST/v2/rover/command

Submit user input and create/continue an authoritative run.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`message`	`string`	User message.
`expectedEpoch`	`number`	Required in strict mode for stale-epoch protection.
`clientEventId`	`string`	Idempotency key for retries and dedup.

Optional Request Fields

Field	Type	Description
`continueRun`	`boolean`	Continue active run. Reserved for ask_user answer continuation in the same task boundary.
`forceNewRun`	`boolean`	Cancel active run and create a fresh run. Normal user sends should set this true.
`expectedSeq`	`number`	Required in strict mode when active run exists.
`requestedMode`	`'act' \| 'planner' \| 'auto'`	Routing preference. `act`/`planner` force backend mode; `auto` uses heuristic routing.
`taskBoundaryId`	`string`	Optional client task boundary identifier.

Success Example

{
  "success": true,
  "data": {
    "runId": "run-uuid",
    "acceptedMode": "act",
    "state": "running",
    "continuePrompt": false,
    "requestedMode": "act",
    "epoch": 5,
    "seq": 1
  }
}

Typed Conflict/Auth Errors

Example 1

{
  "success": false,
  "error": "stale_epoch",
  "data": {
    "sessionId": "visitor-123",
    "runId": "run-uuid",
    "expectedEpoch": 2,
    "currentEpoch": 5,
    "decisionReason": "stale_epoch_retryable",
    "conflict": { "type": "stale_epoch", "currentEpoch": 5, "retryable": true }
  }
}

Example 2

{
  "success": false,
  "error": "active_run_exists",
  "data": {
    "continuePrompt": true,
    "runId": "run-uuid",
    "state": "running",
    "acceptedMode": "act",
    "decisionReason": "active_run_exists",
    "conflict": { "type": "active_run_exists", "retryable": false }
  }
}

Notes

One active run per session is enforced server-side.
No local silent fallback should run when this call is not accepted.

POST/v2/rover/command

Control run lifecycle (`cancel`, `end_task`, `new_task`, `continue`).

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`action`	`'cancel' \| 'end_task' \| 'new_task' \| 'continue'`	Requested run control action.
`expectedEpoch`	`number`	Required in strict mode.
`clientEventId`	`string`	Idempotency key.

Optional Request Fields

Field	Type	Description
`runId`	`string`	Required for `cancel` and `end_task` when active run is not implied.
`expectedSeq`	`number`	Strict stale-seq guard for run-scoped actions.
`reason`	`string`	Audit reason for control action.

Success Example

{
  "success": true,
  "data": {
    "action": "cancel",
    "runId": "run-uuid",
    "currentSeq": 18,
    "projection": { "sessionId": "visitor-123", "epoch": 6, "activeRunId": "", "events": [], "tabs": [] }
  }
}

Notes

Use clientEventId for idempotent retries.
Strict mode requires expectedEpoch and enforces stale rejection before state transitions.

POST/v2/rover/command

Submit navigation/tab intent and receive authoritative policy decision.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`targetUrl \| url`	`string`	Navigation destination.
`expectedEpoch`	`number`	Required in strict mode.
`clientEventId`	`string`	Idempotency key used for silent retry recovery.

Optional Request Fields

Field	Type	Description
`runId`	`string`	Intended active run (defaults to session active run).
`expectedSeq`	`number`	Strict stale-seq guard for run-scoped events.
`currentUrl`	`string`	Current browser URL.
`currentHost/targetHost/isCrossHost/navigationClass`	`hints`	Client hints. Server computes authoritative class regardless.
`logicalTabId`	`string`	Logical tab identifier for run tab graph.
`message`	`string`	Prompt context used for adversarial scoring.
`adversarialScore`	`number`	Optional explicit score (merged with server score).

Success Example

{
  "success": true,
  "data": {
    "decision": "allow_same_tab",
    "decisionReason": "allow_same_tab",
    "decisionHint": "allow_same_tab",
    "notice": "In-scope navigation allowed.",
    "staleRun": false,
    "sessionId": "visitor-123",
    "sessionEpoch": 5,
    "runId": "run-uuid",
    "currentSeq": 3,
    "clientEventId": "evt-uuid",
    "currentHost": "www.rtrvr.ai",
    "targetHost": "rtrvr.ai",
    "isCrossHost": true,
    "crossDomain": false,
    "navigationClass": "cross_host_in_scope",
    "adversarial": { "score": 0, "reasons": [] }
  }
}

Typed Conflict/Auth Errors

Example 1

{
  "success": false,
  "error": "stale_seq",
  "data": {
    "sessionId": "visitor-123",
    "runId": "run-uuid",
    "expectedSeq": 2,
    "currentSeq": 3,
    "currentEpoch": 5,
    "decisionReason": "stale_seq_retryable",
    "clientEventId": "evt-uuid",
    "conflict": { "type": "stale_seq", "currentSeq": 3, "currentEpoch": 5, "retryable": true }
  }
}

Example 2

{
  "success": true,
  "data": {
    "decision": "stale_run",
    "decisionReason": "stale_run",
    "staleRun": true,
    "staleRunReason": "run_terminal",
    "currentRunId": "new-active-run-id",
    "currentActiveRunId": "new-active-run-id",
    "sessionEpoch": 6,
    "decisionHint": "sync_projection_and_ignore"
  }
}

Notes

Server keeps strict stale checks and returns typed 409 envelopes.
SDK should do one silent retry on stale conflicts with same clientEventId.
Stale/missing run is non-fatal via `decision=stale_run` (200).

GET/v2/rover/stream

SSE stream for projection updates and keepalive pings.

Auth: sessionToken (query/body)

Required Request Fields

Field	Type	Description
`sessionId`	`query string`	Session to stream.

Optional Request Fields

Field	Type	Description
`seqAfter`	`number`	Start events after this sequence.

Notes

SSE events: `ready`, `projection`, `ping`, `error`.
Use GET /v2/rover/state as polling fallback.

GET/v2/rover/state

Fetch latest projection (session/run/tabs/events/snapshot metadata).

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`query/body`	Session to fetch.

Optional Request Fields

Field	Type	Description
`seqAfter`	`number`	Return events after this sequence only.

POST/v2/rover/snapshot

Upsert compacted checkpoint/snapshot for resume.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`checkpoint \| payload`	`object`	Compacted snapshot payload.

Optional Request Fields

Field	Type	Description
`visitorId`	`string`	Visitor ID for checkpoint ownership.
`updatedAt`	`number`	Client timestamp in ms.
`version`	`number`	Snapshot schema version.
`seq`	`number`	Last stable seq captured in snapshot.
`compactedPrevSteps`	`unknown[]`	Compacted execution history.
`chatSummary`	`string`	Conversation summary text.
`ttlHours`	`number`	TTL for snapshot document.

POST/v2/rover/context/external

Fetch/act on external context through policy-scoped cloud integrations.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`runId`	`string`	Active run ID.
`url`	`string`	External target URL.
`intent`	`'open_only' \| 'read_context' \| 'act'`	External action intent.
`expectedEpoch`	`number`	Strict stale-epoch guard.
`expectedSeq`	`number`	Strict stale-seq guard.

Optional Request Fields

Field	Type	Description
`logicalTabId`	`string`	Logical external tab id.
`message \| userInput`	`string`	Used for adversarial gating and context.
`timeoutMs`	`number`	External fetch timeout.

Notes

Policy/capability checks are enforced server-side for each call.
High adversarial score can block external context/action even when policy allows it.

POST/v2/rover/telemetry/ingest

Ingest runtime telemetry batches (diagnostics/observability).

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`events`	`unknown[]`	Telemetry event batch (server truncates to safe limits).

Optional Request Fields

Field	Type	Description
`runId`	`string`	Active run association.
`flushReason`	`string`	Flush reason (`interval`, `manual`, etc.).
`pageUrl`	`string`	Current page url.
`sampleRate`	`number`	Client telemetry sample rate.
`sdkVersion`	`string`	SDK version marker.

Notes

This endpoint is automatic; users do not need manual setup for normal onboarding.
Telemetry ingestion does not replace run/state APIs.

Canonical Runtime Envelopes

RoverSuccessResponse<T>

type RoverSuccessResponse<T> = {
  success: true
  data: T
}

RoverErrorResponse

type RoverErrorResponse = {
  success: false
  error: string
  data?: {
    code?: string
    message?: string
    retryable?: boolean
    next_action?: string
    decisionReason?: string
    conflict?: {
      type: 'stale_seq' | 'stale_epoch' | 'active_run_exists'
      currentSeq?: number
      currentEpoch?: number
      retryable: boolean
    }
  }
}

TabEventDecisionResponse

type TabEventDecisionResponse = RoverSuccessResponse<{
  decision: 'allow_same_tab' | 'open_new_tab' | 'block' | 'stale_run'
  decisionReason: 'allow_same_tab' | 'open_new_tab' | 'policy_blocked' | 'stale_run'
  decisionHint: string
  notice?: string
  staleRun: boolean
  staleRunReason?: string
  currentRunId?: string
  currentActiveRunId?: string
  sessionId: string
  sessionEpoch: number
  runId?: string
  currentSeq?: number
  clientEventId?: string
  currentHost?: string
  targetHost?: string
  isCrossHost?: boolean
  crossDomain?: boolean
  navigationClass?: 'same_host_in_scope' | 'cross_host_in_scope' | 'cross_registrable_external'
  adversarial?: { score: number; reasons: string[] }
}>

RunInputResponse

type RunInputResponse = RoverSuccessResponse<{
  runId?: string
  acceptedMode?: 'act' | 'planner'
  state?: 'queued' | 'running' | 'awaiting_user' | 'cancel_requested' | 'cancelled' | 'completed' | 'failed'
  continuePrompt?: boolean
  requestedMode?: 'act' | 'planner' | 'auto'
  epoch?: number
  seq?: number
  routing?: { score: number; reason: string }
}>

RunControlResponse

type RunControlResponse = RoverSuccessResponse<{
  action: 'cancel' | 'end_task' | 'new_task' | 'continue'
  runId?: string
  currentSeq?: number
  projection: SessionProjectionResponse
}>

RunTransitionPayload

type RunTransitionPayload = {
  runId: string
  status: 'queued' | 'running' | 'awaiting_user' | 'cancel_requested' | 'cancelled' | 'completed' | 'failed'
  continuationReason?: 'loop_continue' | 'same_tab_navigation_handoff' | 'awaiting_user'
}

SessionProjectionResponse

type SessionProjectionResponse = {
  sessionId: string
  epoch: number
  activeRunId?: string
  runStatus?: 'queued' | 'running' | 'awaiting_user' | 'cancel_requested' | 'cancelled' | 'completed' | 'failed'
  runMode?: 'act' | 'planner'
  events: Array<{ seq: number; type: string; ts: { _seconds: number; _nanoseconds: number }; data?: Record<string, unknown> }>
  tabs: Array<{ logicalTabId: string; parentLogicalTabId?: string; scope: 'in_scope' | 'external'; status: 'open' | 'blocked' | 'scraped' | 'acted' | 'closed'; url?: string; reason?: string; updatedAt: number }>
  snapshot?: Record<string, unknown>
  snapshotUpdatedAt?: number
}

SSE Event Shape

// GET /v2/rover/stream emits:
// event: ready      data: { sessionId, ts }
// event: projection data: SessionProjectionResponse
// event: ping       data: { ts }
// event: error      data: { message }

Public AgentTask

type AgentTask = {
  id: string
  status: 'pending' | 'waiting_browser' | 'running' | 'input_required' | 'completed' | 'failed' | 'cancelled' | 'expired'
  open?: string
  input?: { questions?: unknown[]; message?: string }
  result?: {
    text?: string
    blocks?: unknown[]
    summary?: string
    observation?: { url?: string; title?: string; host?: string; summary?: string }
    transcript?: { messages?: Array<{ role?: string; text?: string; ts?: number }> }
    artifacts?: Record<string, unknown>[]
  }
}

Public Task Event Shape

// GET /v1/tasks/{id} with SSE or NDJSON emits:
// ready
// status
// step
// tool
// message
// observation
// input
// done
// error
//
// Uniform envelope:
// { id, type, ts, data, chunked?, chunkIndex?, chunkCount?, artifactRef? }

Conflict and Auth Semantics

409 stale_seq when expectedSeq is stale/missing (strict mode) for mutable run-scoped calls.
409 stale_epoch when expectedEpoch is stale/missing (strict mode) for mutable calls.
409 active_run_exists on concurrent run creation races (typed conflict, no generic unknown error).
200 decision='stale_run' for non-fatal stale/missing run in POST /command with type='TAB_EVENT'.
If command-based tab preflight is temporarily unavailable, SDK falls back to local navigation policy instead of hard-blocking navigation.
401 SESSION_TOKEN_EXPIRED, 401 SESSION_TOKEN_INVALID, and 401 BOOTSTRAP_REQUIRED for typed auth flows.

Run Lifecycle and Continuity Semantics

Hard invariant: one active run per session, enforced in backend store/ledger.
requestedMode in POST /command payloads with type='RUN_INPUT': act and planner force backend mode; auto uses heuristic routing.
SDK default: normal user sends are fresh task boundaries with forceNewRun=true (fresh prevSteps + tab scope).
continueRun=true is used only for ask_user answer submission on the active boundary.
Heuristic follow-up chat carryover is cue-only (followupChatLog in worker payloads), and never carries prior task state.
Worker-to-SDK continuation semantics:loop_continue, same_tab_navigation_handoff, awaiting_user.
Navigation outcome semantics:same_tab_scheduled, new_tab_opened, blocked.
Stale seq/epoch races are expected under concurrent navigation/events; SDK should perform one silent retry for command-based tab stale conflicts.

Quick Start Configuration Open Workspace

API Reference

Complete Rover docs for workspace key management, RoverBook Firestore data paths plus private settings callables, public task resources, and browser runtime contracts.

Public task base: https://agent.rtrvr.ai/v1/tasks · Runtime base: https://agent.rtrvr.ai/v2/rover/*

Auth Modes

Firebase ID token for Workspace management APIs (/generateRoverSiteKey, /listRoverSiteKeys, and related config endpoints).
sessionToken (rvrsess_*) for browser runtime calls to /v2/rover/* after session bootstrap.
publicKey (pk_site_*) only for bootstrap exchange on /v2/rover/session/open. Do not send sk_site_* to browser runtime.
task access token for the public task URL returned by /v1/tasks.
Firebase auth + Firestore rules for owner-only RoverBook analytics reads in Workspace, plus Firebase-authenticated callables for private settings such as notification subscriptions.

Public Agent Task API (`/v1/tasks`)

POST/v1/tasks

Create a neutral public task for any Rover-enabled site using a target URL plus a canonical goal or shortcut envelope.

Auth: Anonymous public access token minted in response. Initial create call requires no site public key.

Required Request Fields

Field	Type	Description
`url \| targetUrl`	`string`	Target page URL on a Rover-enabled site.

Optional Request Fields

Field	Type	Description
`goal`	`string`	Canonical natural-language goal for the task.
`prompt`	`string`	Compatibility alias for `goal`.
`shortcutId \| shortcut`	`string`	Exact saved shortcut/capability ID. Use for repeatable flows.
`capabilityId`	`string`	Explicit page/site capability identifier when the caller already knows the target skill.
`args`	`Record<string, unknown>`	Structured task arguments for the selected capability.
`target`	`{ pageId?, route?, capabilityId?, entityId?, formId? }`	Optional page/entity/form targeting hints.
`identity`	`{ userPresent?, signature?, signatureInput?, verificationMethod? }`	Structured caller/user-presence and verification hints.
`execution`	`{ mode?: 'auto' \| 'browser' \| 'cloud'; userPresent?: boolean }`	Requested execution path and user-presence hints.
`accept`	`{ modes?: ('text' \| 'markdown' \| 'json' \| 'observation' \| 'artifacts')[] }`	Requested result modes.
`policy`	`{ confirmation?: 'auto' \| 'required' \| 'site_policy'; crossHost?: 'allow' \| 'same_host' \| 'site_policy' }`	Execution-policy hints that the site may honor or narrow.
`agent`	`{ key?, name?, vendor?, model?, version?, homepage? }`	Optional self-reported visiting-agent metadata used by Rover and RoverBook attribution.

Success Example

{
  "id": "agt_123",
  "task": "https://agent.rtrvr.ai/v1/tasks/agt_123?access=agt_access_...",
  "workflow": "https://agent.rtrvr.ai/v1/workflows/wf_456?access=wf_access_...",
  "open": "https://www.rtrvr.ai/#rover_receipt=rrc_...",
  "browserLink": "https://www.rtrvr.ai/?rover=get+me+the+latest+blog+post#rover_receipt=rrc_...",
  "status": "pending"
}

Notes

Headers drive advanced behavior: `Accept`, `Prefer: wait=15`, `Prefer: execution=auto|browser|cloud`, and `Idempotency-Key`.
Compatibility aliases like `prompt` and `shortcut` still work, but `goal`, `shortcutId`, `capabilityId`, `args`, `target`, `identity`, `execution`, `accept`, and `policy` are the canonical task shape.
If no explicit `agent` object is provided, Rover can still attribute the caller heuristically from `User-Agent`, `Signature-Agent`, `Signature`, `Signature-Input`, and `X-RTRVR-Client-Id`.
Default execution mode is `auto`: browser attach is preferred. `open` is the clean receipt-based browser handoff, while `browserLink` is an optional readable alias when the visible deep link stays within a conservative URL budget.
Use `Prefer: execution=cloud` for guaranteed browserless execution today. Explicit cloud tasks omit browser handoff URLs.

GET/v1/tasks/{id}

Read the canonical public task resource as JSON, SSE, or NDJSON.

Auth: Task access token via `?access=` or `Authorization: Bearer ...`.

Notes

`Accept: application/json` returns latest state or final result.
`Accept: text/event-stream` streams uniform task events.
`Accept: application/x-ndjson` streams the same events in CLI-friendly NDJSON.

POST/v1/tasks/{id}

Continue a task that is waiting on user input.

Auth: Task access token.

Required Request Fields

Field	Type	Description
`input`	`string`	User or agent continuation answer.

DELETE/v1/tasks/{id}

Cancel an in-flight public task.

Auth: Task access token.

POST/v1/tasks/{id}/handoffs

Create a delegated child task on another Rover-enabled site and keep the same workflow lineage.

Auth: Parent task access token.

Required Request Fields

Field	Type	Description
`url`	`string`	Target Rover-enabled site URL.

Optional Request Fields

Field	Type	Description
`instruction`	`string`	Delegation instruction for the target site.
`goal`	`string`	Canonical goal for the delegated child task.
`prompt`	`string`	Compatibility alias for delegated natural-language work.
`shortcutId`	`string`	Exact shortcut to run on the receiving site.
`capabilityId`	`string`	Explicit child-task capability identifier.
`args`	`Record<string, unknown>`	Structured arguments for the delegated capability.
`target`	`{ pageId?, route?, capabilityId?, entityId?, formId? }`	Optional page/entity/form targeting hints on the receiving site.
`identity`	`{ userPresent?, signature?, signatureInput?, verificationMethod? }`	Structured user-presence and verification hints for the child task.
`execution`	`{ mode?: 'auto' \| 'browser' \| 'cloud'; userPresent?: boolean }`	Requested execution path for the child task.
`accept`	`{ modes?: ('text' \| 'markdown' \| 'json' \| 'observation' \| 'artifacts')[] }`	Requested child-task result modes.
`policy`	`{ confirmation?: 'auto' \| 'required' \| 'site_policy'; crossHost?: 'allow' \| 'same_host' \| 'site_policy' }`	Execution-policy hints for the child task.
`agent`	`{ key?, name?, vendor?, model?, version?, homepage? }`	Optional child-task agent identity override. If omitted, the current visiting-agent attribution is inherited.
`contextSummary`	`string`	Structured summary to carry into the child task.
`expectedOutput`	`string`	Describe what the child should return to the parent workflow.

Notes

Delegated child tasks are still ordinary task resources; they simply inherit the parent's workflow lineage.
Receiving sites must enable `siteConfig.aiAccess.allowDelegatedHandoffs` in Workspace.

GET/v1/workflows/{id}

Read the aggregated cross-site workflow resource as JSON, SSE, or NDJSON.

Auth: Workflow access token via `?access=` or `Authorization: Bearer ...`.

Notes

`Accept: application/json` returns the latest workflow snapshot and final result when available.
`Accept: text/event-stream` or `Accept: application/x-ndjson` streams workflow-level lineage across parent and child tasks.

Workspace Management APIs

Used by Rover Workspace for key lifecycle and site config management. These remain Firebase-authenticated control-plane endpoints.

POST/generateRoverSiteKey

Create a new Rover site key and persisted policy/profile.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`siteId \| siteName`	`string`	Existing siteId or a siteName for server-generated siteId.
`label`	`string`	Human-readable key label.
`allowedDomains`	`string[]`	Allowed host/domain patterns (`example.com`, `*.example.com`, `=app.example.com`, or URL-shaped entries normalized to host). In `registrable_domain`, plain `example.com` covers the apex host and subdomains.

Optional Request Fields

Field	Type	Description
`ttlDays`	`number`	Expiration days, use 0 for no expiry.
`environment`	`'production' \| 'development' \| 'test'`	Deployment environment label.
`capabilities`	`{ roverEmbed?, externalWebContextScrape?, cloudAgent?, cloudScrape? }`	Capability profile for this key.
`roverPolicy`	`RoverSitePolicy`	Policy persisted and returned in list/rotate APIs, including `siteMode`, `domainScopeMode`, navigation policy, and shortcut limits.

POST/listRoverSiteKeys

List all Rover site keys for the authenticated workspace user.

Auth: Firebase ID token (Bearer)

POST/updateRoverSiteKeyPolicy

Patch a key's allowed domains, active status, capabilities, and Rover policy.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Target key identifier.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion/update.
`allowedDomains`	`string[]`	Replacement domain pattern set (`example.com`, `*.example.com`, `=app.example.com`, or URL-shaped entries). In `registrable_domain`, plain `example.com` covers the apex host and subdomains.
`active`	`boolean`	Enable/disable key.
`capabilities`	`ApiKeyCapabilities`	Capability patch.
`roverPolicy`	`Partial<RoverSitePolicy>`	Policy patch, including `siteMode`, `domainScopeMode`, and navigation policy updates.

POST/rotateRoverSiteKey

Rotate a key (invalidate old key and issue a new pk_site_* value).

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Key being rotated.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion/update.
`label`	`string`	Optional new label.
`capabilities`	`ApiKeyCapabilities`	Capability patch applied to rotated key.
`roverPolicy`	`Partial<RoverSitePolicy>`	Policy patch applied to the rotated key, including `siteMode`, `domainScopeMode`, and navigation policy.

POST/deleteRoverSite

Delete a Rover site, all owned site keys, and Rover-owned config while retaining historical RoverBook analytics.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`siteId`	`string`	Rover site identifier to delete.

Notes

Deletes all owned Rover site keys for the site, plus Rover-owned site config and RoverBook owner settings/questions.
Historical RoverBook analytics, visits, reviews, interviews, notes, posts, and aggregates are retained.
Webflow-managed sites are detached from the live managed Webflow install before Rover-owned resources are removed.

POST/getRoverSiteConfig

Read cloud shortcuts, business type, sparse experience overrides, greeting, voice, AI access, discovery, and page-capture config for a key/site.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Site key ID.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion.

POST/upsertRoverSiteConfig

Write cloud shortcuts, business type, sparse experience overrides, greeting, voice, AI access, discovery, and page-capture config.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Site key ID.

Optional Request Fields

Field	Type	Description
`siteId`	`string`	Optional site assertion.
`siteConfig.shortcuts`	`RoverSiteShortcut[]`	Persisted shortcuts/journeys.
`siteConfig.businessType`	`RoverBusinessType`	Tunes runtime placeholder hints and generated quick actions when the site has few explicit shortcuts.
`siteConfig.experience`	`RoverSiteExperience \| null`	Sparse Rover V3 surface overrides only. Runtime defaults still own the seed, centered stage, and focus stream when a field is omitted. Use null to clear overrides.
`siteConfig.greeting`	`{ text?, delay?, duration?, disabled? } \| null`	Greeting config.
`siteConfig.voice`	`{ enabled?: boolean; language?: string; autoStopMs?: number } \| null`	Browser dictation config. autoStopMs is the post-speech silence window in milliseconds. Use null to clear persisted voice settings.
`siteConfig.aiAccess`	`RoverSiteAiAccess \| null`	Public launch/runtime guardrails for ATP task entry, browser deep links, cloud browser use, and delegated handoffs.
`siteConfig.agentDiscovery`	`RoverSiteAgentDiscovery \| null`	Persisted public discovery policy, well-known artifact paths, host-surface binding, and seed/presence-led display settings. Use null to clear.
`siteConfig.pageConfig`	`RoverPageCaptureConfig \| null`	Sparse per-site page-capture overrides. Use null to clear overrides.

POST/updateApiKeyCapabilities

Patch capability flags for user or site keys.

Auth: Firebase ID token (Bearer)

Required Request Fields

Field	Type	Description
`keyId`	`string`	Target key identifier.
`capabilities`	`ApiKeyCapabilities`	Boolean capability patch fields.

RoverBook Workspace Data Paths

READFirestore: roverbook_* collections

Owner-facing RoverBook analytics now read Firestore directly under ownership rules instead of calling a backend read facade.

Auth: Firebase auth + Firestore rules (`ownerUid` scoped).

Notes

Workspace reads `roverbook_sites`, `roverbook_visits`, `roverbook_scores`, `roverbook_reviews`, `roverbook_interviews`, `roverbook_questions`, `roverbook_notes`, and `roverbook_posts` directly.
Signed public RoverBook GET routes still exist for runtime/agent surfaces used by `@rover/roverbook`.

POSTgetRoverBookNotificationSettings

Read the private RoverBook setup payload for a site, including interview questions and masked webhook subscriptions.

Auth: Firebase ID token (owner only).

Required Request Fields

Field	Type	Description
`siteId`	`string`	Owned Rover site identifier.

POSTupsertRoverBookNotificationSettings

Write RoverBook interview questions and webhook subscriptions for a site.

Auth: Firebase ID token (owner only).

Required Request Fields

Field	Type	Description
`siteId`	`string`	Owned Rover site identifier.

Optional Request Fields

Field	Type	Description
`enabled`	`boolean`	Enable or disable RoverBook outbound notifications for the site.
`webhooks`	`RoverBookWebhookSubscription[]`	Private per-site webhook subscription list.
`interviewQuestions`	`string[]`	Per-site RoverBook interview prompts.

Notes

Webhook secrets are masked on read and preserved unless explicitly replaced.
These settings are stored privately by owner + site and are not exposed through public embed config.

POSTupsertRoverBookQuestions

Write just the RoverBook interview questions for a site.

Auth: Firebase ID token (owner only).

Required Request Fields

Field	Type	Description
`siteId`	`string`	Owned Rover site identifier.
`interviewQuestions`	`string[]`	Per-site RoverBook interview prompts.

Rover Runtime APIs (`/v2/rover/*`)

Server-authoritative runtime contract for embedded Rover. All run-state changes are keyed bysessionId + runId + epoch + seq.

POST/v2/rover/session/open

Bootstrap or refresh a browser runtime session and return the initial projection.

Auth: Body token. Preferred: bootstrapToken/publicKey (pk_site_*). Optional: sessionToken refresh path.

Required Request Fields

Field	Type	Description
`siteId`	`string`	Workspace site identifier.
`host \| url`	`string`	Current host/url for domain and policy checks.
`bootstrapToken \| publicKey`	`string`	Public Rover key (pk_site_*). Required when no valid sessionToken is supplied.

Optional Request Fields

Field	Type	Description
`sessionId`	`string`	Client session identifier (auto-generated if omitted).
`sessionToken`	`string`	Current rvrsess_* token. If invalid/expired and bootstrap token exists, server falls back to bootstrap flow.

Success Example

{
  "success": true,
  "data": {
    "sessionId": "visitor-123",
    "sessionToken": "rvrsess_...",
    "sessionTokenExpiresAt": 1771484403000,
    "streamToken": "rvrsess_...",
    "epoch": 5,
    "capabilities": { "roverEmbed": true },
    "policy": {
      "domainScopeMode": "registrable_domain",
      "externalNavigationPolicy": "open_new_tab_notice",
      "crossHostPolicy": "same_tab",
      "enableExternalWebContext": true,
      "externalScrapeMode": "on_demand",
      "externalAllowDomains": [],
      "externalDenyDomains": []
    },
	    "projection": { "sessionId": "visitor-123", "epoch": 5, "events": [], "tabs": [] },
	    "siteConfig": {
	      "shortcuts": [],
	      "businessType": "saas",
	      "experience": {
	        "presence": { "ctaText": "Do it with Rover" },
	        "inputs": { "voice": false }
	      },
	      "greeting": { "text": "Welcome" },
	      "voice": { "enabled": true, "language": "en-US", "autoStopMs": 2600 },
	      "aiAccess": { "enabled": true },
	      "agentDiscovery": {
        "enabled": true,
        "preferExecution": "auto",
        "roverSiteUrl": "/.well-known/rover-site.json",
        "agentCardUrl": "/.well-known/agent-card.json",
        "discoverySurface": {
          "mode": "beacon",
          "branding": "site",
          "hostSurface": "auto",
          "actionReveal": "click",
          "beaconLabel": "Use AI"
        }
      },
      "pageConfig": { "disableAutoScroll": true, "adaptiveSettleMaxWaitMs": 480 }
    },
    "sseUrl": "https://agent.rtrvr.ai/v2/rover/stream?..."
  }
}

Typed Conflict/Auth Errors

Example 1

{
  "success": false,
  "error": "SESSION_TOKEN_EXPIRED",
  "data": {
    "code": "SESSION_TOKEN_EXPIRED",
    "message": "Token expired",
    "retryable": true,
    "next_action": "Pass bootstrapToken/publicKey (pk_site_*) to /v2/rover/session/open to re-bootstrap the session."
  }
}

Example 2

{
  "success": false,
  "error": "BOOTSTRAP_REQUIRED",
  "data": {
    "code": "BOOTSTRAP_REQUIRED",
    "message": "bootstrapToken/publicKey (pk_site_*) is required.",
    "retryable": false,
    "next_action": "Provide a valid pk_site_* key in bootstrapToken/publicKey."
  }
}

Notes

Server mints short-lived sessionToken (~10 min).
Browser runtime should not send long-lived bearer keys.

POST/v2/rover/session/open

Refresh an existing rvrsess_* token and SSE URL.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Session to refresh (defaults to token sid).

Notes

Use when token is near expiry or after SSE auth failures.

POST/v2/rover/command

Submit user input and create/continue an authoritative run.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`message`	`string`	User message.
`expectedEpoch`	`number`	Required in strict mode for stale-epoch protection.
`clientEventId`	`string`	Idempotency key for retries and dedup.

Optional Request Fields

Field	Type	Description
`continueRun`	`boolean`	Continue active run. Reserved for ask_user answer continuation in the same task boundary.
`forceNewRun`	`boolean`	Cancel active run and create a fresh run. Normal user sends should set this true.
`expectedSeq`	`number`	Required in strict mode when active run exists.
`requestedMode`	`'act' \| 'planner' \| 'auto'`	Routing preference. `act`/`planner` force backend mode; `auto` uses heuristic routing.
`taskBoundaryId`	`string`	Optional client task boundary identifier.

Success Example

{
  "success": true,
  "data": {
    "runId": "run-uuid",
    "acceptedMode": "act",
    "state": "running",
    "continuePrompt": false,
    "requestedMode": "act",
    "epoch": 5,
    "seq": 1
  }
}

Typed Conflict/Auth Errors

Example 1

{
  "success": false,
  "error": "stale_epoch",
  "data": {
    "sessionId": "visitor-123",
    "runId": "run-uuid",
    "expectedEpoch": 2,
    "currentEpoch": 5,
    "decisionReason": "stale_epoch_retryable",
    "conflict": { "type": "stale_epoch", "currentEpoch": 5, "retryable": true }
  }
}

Example 2

{
  "success": false,
  "error": "active_run_exists",
  "data": {
    "continuePrompt": true,
    "runId": "run-uuid",
    "state": "running",
    "acceptedMode": "act",
    "decisionReason": "active_run_exists",
    "conflict": { "type": "active_run_exists", "retryable": false }
  }
}

Notes

One active run per session is enforced server-side.
No local silent fallback should run when this call is not accepted.

POST/v2/rover/command

Control run lifecycle (`cancel`, `end_task`, `new_task`, `continue`).

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`action`	`'cancel' \| 'end_task' \| 'new_task' \| 'continue'`	Requested run control action.
`expectedEpoch`	`number`	Required in strict mode.
`clientEventId`	`string`	Idempotency key.

Optional Request Fields

Field	Type	Description
`runId`	`string`	Required for `cancel` and `end_task` when active run is not implied.
`expectedSeq`	`number`	Strict stale-seq guard for run-scoped actions.
`reason`	`string`	Audit reason for control action.

Success Example

{
  "success": true,
  "data": {
    "action": "cancel",
    "runId": "run-uuid",
    "currentSeq": 18,
    "projection": { "sessionId": "visitor-123", "epoch": 6, "activeRunId": "", "events": [], "tabs": [] }
  }
}

Notes

Use clientEventId for idempotent retries.
Strict mode requires expectedEpoch and enforces stale rejection before state transitions.

POST/v2/rover/command

Submit navigation/tab intent and receive authoritative policy decision.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`targetUrl \| url`	`string`	Navigation destination.
`expectedEpoch`	`number`	Required in strict mode.
`clientEventId`	`string`	Idempotency key used for silent retry recovery.

Optional Request Fields

Field	Type	Description
`runId`	`string`	Intended active run (defaults to session active run).
`expectedSeq`	`number`	Strict stale-seq guard for run-scoped events.
`currentUrl`	`string`	Current browser URL.
`currentHost/targetHost/isCrossHost/navigationClass`	`hints`	Client hints. Server computes authoritative class regardless.
`logicalTabId`	`string`	Logical tab identifier for run tab graph.
`message`	`string`	Prompt context used for adversarial scoring.
`adversarialScore`	`number`	Optional explicit score (merged with server score).

Success Example

{
  "success": true,
  "data": {
    "decision": "allow_same_tab",
    "decisionReason": "allow_same_tab",
    "decisionHint": "allow_same_tab",
    "notice": "In-scope navigation allowed.",
    "staleRun": false,
    "sessionId": "visitor-123",
    "sessionEpoch": 5,
    "runId": "run-uuid",
    "currentSeq": 3,
    "clientEventId": "evt-uuid",
    "currentHost": "www.rtrvr.ai",
    "targetHost": "rtrvr.ai",
    "isCrossHost": true,
    "crossDomain": false,
    "navigationClass": "cross_host_in_scope",
    "adversarial": { "score": 0, "reasons": [] }
  }
}

Typed Conflict/Auth Errors

Example 1

{
  "success": false,
  "error": "stale_seq",
  "data": {
    "sessionId": "visitor-123",
    "runId": "run-uuid",
    "expectedSeq": 2,
    "currentSeq": 3,
    "currentEpoch": 5,
    "decisionReason": "stale_seq_retryable",
    "clientEventId": "evt-uuid",
    "conflict": { "type": "stale_seq", "currentSeq": 3, "currentEpoch": 5, "retryable": true }
  }
}

Example 2

{
  "success": true,
  "data": {
    "decision": "stale_run",
    "decisionReason": "stale_run",
    "staleRun": true,
    "staleRunReason": "run_terminal",
    "currentRunId": "new-active-run-id",
    "currentActiveRunId": "new-active-run-id",
    "sessionEpoch": 6,
    "decisionHint": "sync_projection_and_ignore"
  }
}

Notes

Server keeps strict stale checks and returns typed 409 envelopes.
SDK should do one silent retry on stale conflicts with same clientEventId.
Stale/missing run is non-fatal via `decision=stale_run` (200).

GET/v2/rover/stream

SSE stream for projection updates and keepalive pings.

Auth: sessionToken (query/body)

Required Request Fields

Field	Type	Description
`sessionId`	`query string`	Session to stream.

Optional Request Fields

Field	Type	Description
`seqAfter`	`number`	Start events after this sequence.

Notes

SSE events: `ready`, `projection`, `ping`, `error`.
Use GET /v2/rover/state as polling fallback.

GET/v2/rover/state

Fetch latest projection (session/run/tabs/events/snapshot metadata).

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`query/body`	Session to fetch.

Optional Request Fields

Field	Type	Description
`seqAfter`	`number`	Return events after this sequence only.

POST/v2/rover/snapshot

Upsert compacted checkpoint/snapshot for resume.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`checkpoint \| payload`	`object`	Compacted snapshot payload.

Optional Request Fields

Field	Type	Description
`visitorId`	`string`	Visitor ID for checkpoint ownership.
`updatedAt`	`number`	Client timestamp in ms.
`version`	`number`	Snapshot schema version.
`seq`	`number`	Last stable seq captured in snapshot.
`compactedPrevSteps`	`unknown[]`	Compacted execution history.
`chatSummary`	`string`	Conversation summary text.
`ttlHours`	`number`	TTL for snapshot document.

POST/v2/rover/context/external

Fetch/act on external context through policy-scoped cloud integrations.

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`runId`	`string`	Active run ID.
`url`	`string`	External target URL.
`intent`	`'open_only' \| 'read_context' \| 'act'`	External action intent.
`expectedEpoch`	`number`	Strict stale-epoch guard.
`expectedSeq`	`number`	Strict stale-seq guard.

Optional Request Fields

Field	Type	Description
`logicalTabId`	`string`	Logical external tab id.
`message \| userInput`	`string`	Used for adversarial gating and context.
`timeoutMs`	`number`	External fetch timeout.

Notes

Policy/capability checks are enforced server-side for each call.
High adversarial score can block external context/action even when policy allows it.

POST/v2/rover/telemetry/ingest

Ingest runtime telemetry batches (diagnostics/observability).

Auth: sessionToken

Required Request Fields

Field	Type	Description
`sessionId`	`string`	Target session.
`events`	`unknown[]`	Telemetry event batch (server truncates to safe limits).

Optional Request Fields

Field	Type	Description
`runId`	`string`	Active run association.
`flushReason`	`string`	Flush reason (`interval`, `manual`, etc.).
`pageUrl`	`string`	Current page url.
`sampleRate`	`number`	Client telemetry sample rate.
`sdkVersion`	`string`	SDK version marker.

Notes

This endpoint is automatic; users do not need manual setup for normal onboarding.
Telemetry ingestion does not replace run/state APIs.

Canonical Runtime Envelopes

RoverSuccessResponse<T>

type RoverSuccessResponse<T> = {
  success: true
  data: T
}

RoverErrorResponse

type RoverErrorResponse = {
  success: false
  error: string
  data?: {
    code?: string
    message?: string
    retryable?: boolean
    next_action?: string
    decisionReason?: string
    conflict?: {
      type: 'stale_seq' | 'stale_epoch' | 'active_run_exists'
      currentSeq?: number
      currentEpoch?: number
      retryable: boolean
    }
  }
}

TabEventDecisionResponse

type TabEventDecisionResponse = RoverSuccessResponse<{
  decision: 'allow_same_tab' | 'open_new_tab' | 'block' | 'stale_run'
  decisionReason: 'allow_same_tab' | 'open_new_tab' | 'policy_blocked' | 'stale_run'
  decisionHint: string
  notice?: string
  staleRun: boolean
  staleRunReason?: string
  currentRunId?: string
  currentActiveRunId?: string
  sessionId: string
  sessionEpoch: number
  runId?: string
  currentSeq?: number
  clientEventId?: string
  currentHost?: string
  targetHost?: string
  isCrossHost?: boolean
  crossDomain?: boolean
  navigationClass?: 'same_host_in_scope' | 'cross_host_in_scope' | 'cross_registrable_external'
  adversarial?: { score: number; reasons: string[] }
}>

RunInputResponse

type RunInputResponse = RoverSuccessResponse<{
  runId?: string
  acceptedMode?: 'act' | 'planner'
  state?: 'queued' | 'running' | 'awaiting_user' | 'cancel_requested' | 'cancelled' | 'completed' | 'failed'
  continuePrompt?: boolean
  requestedMode?: 'act' | 'planner' | 'auto'
  epoch?: number
  seq?: number
  routing?: { score: number; reason: string }
}>

RunControlResponse

type RunControlResponse = RoverSuccessResponse<{
  action: 'cancel' | 'end_task' | 'new_task' | 'continue'
  runId?: string
  currentSeq?: number
  projection: SessionProjectionResponse
}>

RunTransitionPayload

type RunTransitionPayload = {
  runId: string
  status: 'queued' | 'running' | 'awaiting_user' | 'cancel_requested' | 'cancelled' | 'completed' | 'failed'
  continuationReason?: 'loop_continue' | 'same_tab_navigation_handoff' | 'awaiting_user'
}

SessionProjectionResponse

type SessionProjectionResponse = {
  sessionId: string
  epoch: number
  activeRunId?: string
  runStatus?: 'queued' | 'running' | 'awaiting_user' | 'cancel_requested' | 'cancelled' | 'completed' | 'failed'
  runMode?: 'act' | 'planner'
  events: Array<{ seq: number; type: string; ts: { _seconds: number; _nanoseconds: number }; data?: Record<string, unknown> }>
  tabs: Array<{ logicalTabId: string; parentLogicalTabId?: string; scope: 'in_scope' | 'external'; status: 'open' | 'blocked' | 'scraped' | 'acted' | 'closed'; url?: string; reason?: string; updatedAt: number }>
  snapshot?: Record<string, unknown>
  snapshotUpdatedAt?: number
}

SSE Event Shape

// GET /v2/rover/stream emits:
// event: ready      data: { sessionId, ts }
// event: projection data: SessionProjectionResponse
// event: ping       data: { ts }
// event: error      data: { message }

Public AgentTask

type AgentTask = {
  id: string
  status: 'pending' | 'waiting_browser' | 'running' | 'input_required' | 'completed' | 'failed' | 'cancelled' | 'expired'
  open?: string
  input?: { questions?: unknown[]; message?: string }
  result?: {
    text?: string
    blocks?: unknown[]
    summary?: string
    observation?: { url?: string; title?: string; host?: string; summary?: string }
    transcript?: { messages?: Array<{ role?: string; text?: string; ts?: number }> }
    artifacts?: Record<string, unknown>[]
  }
}

Public Task Event Shape

// GET /v1/tasks/{id} with SSE or NDJSON emits:
// ready
// status
// step
// tool
// message
// observation
// input
// done
// error
//
// Uniform envelope:
// { id, type, ts, data, chunked?, chunkIndex?, chunkCount?, artifactRef? }

Conflict and Auth Semantics

409 stale_seq when expectedSeq is stale/missing (strict mode) for mutable run-scoped calls.
409 stale_epoch when expectedEpoch is stale/missing (strict mode) for mutable calls.
409 active_run_exists on concurrent run creation races (typed conflict, no generic unknown error).
200 decision='stale_run' for non-fatal stale/missing run in POST /command with type='TAB_EVENT'.
If command-based tab preflight is temporarily unavailable, SDK falls back to local navigation policy instead of hard-blocking navigation.
401 SESSION_TOKEN_EXPIRED, 401 SESSION_TOKEN_INVALID, and 401 BOOTSTRAP_REQUIRED for typed auth flows.

Run Lifecycle and Continuity Semantics

Hard invariant: one active run per session, enforced in backend store/ledger.
requestedMode in POST /command payloads with type='RUN_INPUT': act and planner force backend mode; auto uses heuristic routing.
SDK default: normal user sends are fresh task boundaries with forceNewRun=true (fresh prevSteps + tab scope).
continueRun=true is used only for ask_user answer submission on the active boundary.
Heuristic follow-up chat carryover is cue-only (followupChatLog in worker payloads), and never carries prior task state.
Worker-to-SDK continuation semantics:loop_continue, same_tab_navigation_handoff, awaiting_user.
Navigation outcome semantics:same_tab_scheduled, new_tab_opened, blocked.
Stale seq/epoch races are expected under concurrent navigation/events; SDK should perform one silent retry for command-based tab stale conflicts.

Quick Start Configuration Open Workspace