title	Node.js API

Node.js API

This page is for developers who drive devices from their own Node.js code. Use createAgentDeviceClient() for typed, deterministic device automation instead of shelling out to the CLI.

Building an agent? Start with the dedicated AI SDK or Eve integration. Building a bridge, remote daemon, or device-cloud backend that owns device access or command dispatch? See Build an integration.

Runnable examples

The repository includes runnable, typechecked Node.js examples. They import the same published agent-device/* entry points your code uses:

Example	Demonstrates
client-session.ts	Open, snapshot, interact, handle typed errors, and always close the session
contracts-result.ts	Consume snapshot results with helpers from agent-device/contracts

Examples for custom transports and Metro are on Build an integration.

Each example typechecks against the SDK through its own tsconfig.json. Build the package with pnpm build, then run an example with Node:

node --experimental-strip-types examples/sdk/client-session.ts

API reference

Entry points for driving devices from Node.js:

• agent-device
  • createAgentDeviceClient(options?)
  • ensureDaemon(options?)
  • createLocalArtifactAdapter(options?)
  • AppError, isAgentDeviceError(error), normalizeAgentDeviceError(error)
  • isUnreadableCaptureContentError(error): true when an Android helper capture found no readable app content yet (empty-helper-output, system-window-only, content-poor-app-window), for example while the app is still loading. Polling callers may keep polling. False for every other platform, backend, and mechanism failure.
  • centerOfRect(rect)
  • types: AgentDeviceClient, AgentDeviceClientConfig, AgentDeviceDevice
  • types: DaemonLifecycle, EnsureDaemonOptions, DaemonConnection, DaemonIdentity
  • types: every option and result type of the client's methods, among them AppOpenOptions, AppOpenResult, CaptureSnapshotOptions, CaptureSnapshotResult, CaptureScreenshotResult, PressOptions
  • types: SnapshotNode, RawSnapshotNode, SnapshotState, Rect, Point
  • types: NormalizedError, AppErrorCode, KnownAppErrorCode, AppErrorDetails, ErrorCause
• agent-device/io
  • createLocalArtifactAdapter(options?)
  • types: ArtifactAdapter, ArtifactDescriptor, CreateTempFileOptions, FileInputRef, FileOutputRef, LocalArtifactAdapterOptions, OutputVisibility, ReserveOutputOptions, ReservedOutputFile, ResolveInputOptions, ResolvedInputFile, TemporaryFile
• agent-device/ai-sdk
  • createAgentDeviceTools(options)
  • types: AgentDeviceToolSet, AgentDeviceTools, CreateAgentDeviceToolsOptions

Basic usage

This example opens an app, takes a snapshot, interacts, handles typed errors, and always closes the session. Run it from examples/sdk/client-session.ts.

client.devices.capabilities() returns { device, availableCommands }, using the same capability matrix as the CLI. Use it to check which command names the selected target supports.

For direct iOS simulator app launches, client.apps.open({ app, platform: 'ios', launchConsole: './artifacts/app.console.log' }) captures launch-time stdout/stderr. The option mirrors open --launch-console and is not valid for URL opens or non-simulator targets.

client.sessions.stateDir() mirrors session state-dir and returns the resolved daemon state directory without starting or contacting the daemon. Pass { stateDir } to resolve an explicit override the same way the CLI resolves --state-dir.

client.sessions.artifacts({ provider, providerSessionId }) mirrors artifacts --provider ... --provider-session ... and returns provider-hosted cloudArtifacts. Use it for BrowserStack, AWS Device Farm, or TestMu AI session videos/logs after a cloud session has stopped; client.sessions.close() returns the ID as closed.provider?.providerSessionId. Omit providerSessionId when an embedding host has registered a provider runtime that can infer the active lease. Limrun does not expose provider artifacts through this command.

const result = await client.sessions.artifacts({
  provider: 'aws-device-farm',
  providerSessionId: 'arn:aws:devicefarm:us-west-2:123:session/project/session/00000',
});

if ('cloudArtifacts' in result) {
  for (const artifact of result.cloudArtifacts) {
    console.log(artifact.kind, artifact.name, artifact.url);
  }
}

Daemon lifecycle

A client talks to one local daemon per state directory. daemonLifecycle says who owns that daemon.

Mode	What the client does with the daemon
'managed' (default)	Starts a missing daemon, replaces one that does not match the client, and retires an unresponsive one after a timeout.
'attach'	Connects to the running daemon. Never starts, replaces or stops one, even after a timeout.

'managed' stays the default until 0.22. Use 'attach' when your code, or another process, owns the daemon, and call ensureDaemon() to start or replace it deliberately:

import { createAgentDeviceClient, ensureDaemon } from 'agent-device';

const stateDir = './.my-app/agent-device-state';

await ensureDaemon({
  stateDir,
  onStart: (reason, previous) => console.log(`started a daemon: ${reason}`, previous),
});
const client = createAgentDeviceClient({ stateDir, daemonLifecycle: 'attach' });

ensureDaemon() starts a daemon when none runs. It replaces a daemon that does not match the client only if that daemon holds no work: no request in flight, no session, no repair session and no live lease. A resumable upload still streaming and a host human-control hold that names no lease are not counted, so a replacement can end them. Otherwise it fails and leaves the daemon alone. Two cases retire a daemon without that proof, because it cannot be asked: a registered daemon that answers nothing on any transport, and one that times out on the quiesce request and then fails repeated liveness probes. It returns { stateDir, daemon, started }; daemon holds the pid, version and code identity, and no credentials. onStart runs when the call started a daemon: 'incompatible' when it replaced a daemon that agreed to leave, with that daemon as previous, and 'missing' otherwise. A 'missing' start that also has a previous replaced a registered daemon that answered nothing.

These failures carry a typed error.details.reason:

Reason	Meaning
daemon-not-running	An 'attach' client found no daemon to connect to. Call ensureDaemon().
daemon-incompatible	The running daemon does not match the client. details.daemon and details.client hold both identities. Stop it with agent-device daemon stop --state-dir <path>, then call ensureDaemon(). A daemon from a release that predates quiescing cannot be asked to leave, so ensureDaemon() reports it the same way.
daemon-busy	ensureDaemon() found an incompatible daemon that still holds work. details counts inFlightRequests, sessions, repairSessions and leases. Wait for them to end and retry.

ensureDaemon() only ever manages the local daemon on stateDir. A client with daemonBaseUrl never starts or stops a local daemon in either mode, and daemonLifecycle does not change daemonTransport or daemonServerMode.

Device cloud sessions

You drive Limrun, BrowserStack, AWS Device Farm, and TestMu AI devices with the same typed client methods. Use the CLI connect flow to persist connection state locally. Pass provider settings in the client config when your Node.js integration already owns credentials and provider selectors.

import { createAgentDeviceClient } from 'agent-device';

const client = createAgentDeviceClient({
  leaseProvider: 'browserstack',
  providerOsVersion: '14.0',
  providerApp: 'bs://app-id',
  // Optional hosted device features, applied when the session is created.
  providerDeviceOrientation: 'portrait',
  providerTimezone: 'New_York',
});

await client.apps.open({ app: 'com.example.app', platform: 'android', device: 'Google Pixel 8' });
await client.capture.snapshot({ interactiveOnly: true });
const closed = await client.sessions.close();

apps.open also returns a response-level selection record describing whether the target came from an explicit selector, an existing session, one local booted/bootable candidate, the one booted simulator with the app installed, or one provider-owned candidate. Ambiguous requests fail with structured retry selectors instead of silently retargeting.

To fetch provider-hosted video and log URLs after close, pass closed.provider?.providerSessionId to client.sessions.artifacts(). See the BrowserStack, AWS Device Farm, TestMu AI, and Limrun guides for provider-specific setup.

Web sessions

Pass platform: 'web' to drive a browser session with the same command methods. Web automation requires Node 24+. No client method sets up the managed web backend: run agent-device web setup before first use, in the same state directory the client uses. Run agent-device web doctor to check backend health.

await client.apps.open({ url: 'https://example.com', platform: 'web' });
await client.capture.snapshot({ platform: 'web', interactiveOnly: true });
await client.interactions.fill({ platform: 'web', ref: '@e12', text: '[email protected]' });
await client.command.wait({ platform: 'web', text: 'Welcome' });
await client.observability.network({ platform: 'web', include: 'headers' });
await client.observability.audio({
  platform: 'web',
  action: 'probe',
  probeAction: 'start',
  durationMs: 10_000,
  bucketMs: 1_000,
});
await client.sessions.close();

MCP tools use the same command contracts, so they can also target platform: 'web' after setup; setup and doctor stay CLI-only. Web network inspection returns the standard network result shape without request or response bodies.

Command methods

Use client.command.<method>() for command-level device actions. These calls go through the same daemon transport as the higher-level client methods, so they carry session metadata and tenant/run/lease fields, return normalized daemon errors, and handle remote artifacts.

Results are daemon-shaped objects with typed known fields and follow the CLI's command semantics.

A failed interaction rejects with the same error the CLI prints. Read error.details.dispatched before you retry; Retry after a failed command explains the values.

Every client call that dispatches to the daemon accepts signal?: AbortSignal to cancel that one call:

const controller = new AbortController();
await client.interactions.press({ ref: '@e12', signal: controller.signal });

With the built-in transport, a signal that is already aborted rejects the call without sending anything (error.details.dispatched: 'no'). Aborting while the request is in flight closes that request's connection, the daemon marks the request canceled, and the promise rejects with the typed canceled-request error (error.details.reason: 'request_canceled', error.details.dispatched: 'unknown'). A custom transport receives the signal on its context and may cancel differently. The daemon and the session stay alive for other requests. On iOS, a command that already reached the device keeps running there, so the daemon holds the session while the device reports it running. Your promise still rejects at once. The hold ends when the device reports the command finished, when the command's own timeout runs out, or when the device cannot confirm the command (it never received it, or the status check fails). Only in the first case is the canceled work known to be over before the session's next request starts. An abort is not a timeout: it never triggers the runner cleanup or daemon reset that a timeout does. Cancellation covers only the daemon request, so a response-artifact download already underway is not stopped, and a canceled one-shot replay still runs the existing cleanup that can tear down a daemon this client started.

await client.command.wait({
  text: 'Continue',
  timeoutMs: 5_000,
});

await client.command.keyboard({
  action: 'dismiss',
});

await client.command.clipboard({
  action: 'write',
  text: 'hello from Node',
});

await client.command.back({
  mode: 'system',
});

await client.command.tvRemote({
  platform: 'android',
  target: 'tv',
  button: 'down',
});

await client.command.tvRemote({
  platform: 'ios',
  target: 'tv',
  button: 'select',
});

await client.command.tvRemote({
  platform: 'vega',
  target: 'tv',
  serial: 'VirtualDevice',
  button: 'select',
  durationMs: 900,
});

await client.command.appSwitcher();
await client.command.actionButton();
await client.command.fold({ pose: 'open' });
await client.command.telephony({ action: 'call', number: '5551234' });
await client.command.telephony({ action: 'sms', number: '5551234', text: 'Your code is 1234' });
await client.command.fold({
  keyframes: [
    { atMs: 0, angle: 0 },
    { atMs: 1667, angle: 160 },
    { atMs: 3333, angle: 100 },
    { atMs: 5000, angle: 180 },
  ],
});

fold accepts either pose or keyframes. Keyframes use linear interpolation at roughly 60 updates per second; repeat an angle to hold it. Timestamps must start at zero and increase strictly, with 2–64 frames and a final timestamp no greater than 60,000ms. Angles must be finite and between 0° and 180°. The final timestamp bounds motion, excluding helper preparation and final hinge verification. A custom final angle is verified within 0.5°; interior angles must also settle. Cancellation stops the motion at its current angle. Re-snapshot afterwards, including after interrupted motion. Android emulators accept pose only.

press, click, and longpress do not wait for a missing target to appear. When a screen is still loading, wait for its landmark before interacting:

await client.command.wait({ text: 'Continue', timeoutMs: 5_000 });
await client.interactions.press({ selector: 'label="Continue"' });

If the wait times out, check the selector and app state with a fresh snapshot. A covered, off-screen, or ambiguous target still fails the interaction.

Vega OS client support is VVD-only and covers device discovery, app open/close, back, home, and tvRemote. Physical Fire TV, capture, selector, install, logging, and performance methods report unsupported for Vega targets.

Supported command methods:

• wait
• alert
• appState
• back
• home
• orientation
• appSwitcher
• actionButton
• fold
• telephony
• keyboard
• clipboard
• tvRemote
• reactNative
• doctor
• prepare
• viewport

rotate() is a deprecated alias for orientation(); use orientation().

Domain client methods:

• client.devices.list(), capabilities(), boot(), shutdown()
• client.sessions.list(), stateDir(), close(), saveScript(), artifacts()
• client.apps.install(), reinstall(), uninstall(), installFromSource(), list(), open(), close(), push(), triggerEvent()
• client.materializations.release()
• client.leases.allocate(), heartbeat(), release(). Pass retainOnClose: true to allocate() when you release the lease yourself; session close then leaves it active until release(), expiry, or daemon shutdown. The returned lease has retainOnClose: true only when the daemon honored it. Keep heartbeating a retained lease between sessions: it expires after its ttlMs like any other lease. Pass leaseId (16 to 128 hex characters) to allocate() to name a new lease with an id the daemon has never issued, or to get back your live lease with that id when you retry with the same tenant, run, client, lease backend, and provider (and no device, or the leased one), then use the returned leaseId: a run that already holds a lease gets that one back, and an older daemon mints its own.
• client.metro.prepare(), reload()
• client.capture.snapshot(), screenshot(), diff()
• client.interactions.click(), press(), longPress(), swipe(), pan(), drag(), fling(), swipeGesture(), focus(), type(), fill(), scroll(), pinch(), rotateGesture(), transformGesture(), get(), is(), find()
• client.replay.run() and client.replay.test()
• client.batch.run()
• client.observability.perf(options), logs(), events(), network(), and audio()
• client.debug.symbols()
• client.recording.record() and client.recording.trace()
• client.settings.update()

client.devices.list() returns AgentDeviceDevice entries. Their optional model and osVersion fields describe the hardware and OS when discovery reports them; see Device discovery for the sources.

client.capture.snapshot() carries an optional viewport: { width, height } beside nodes: the box those rects are measured in, in the same coordinate space and orientation. Scale and clip against it instead of inferring the screen from the largest rect. When no box was measured, the field is absent rather than zero; see Coordinates and viewport for what each producer answers with.

It also carries an optional presentedSurface: 'presented' | 'absent' | 'unknown' on iOS captures: whether the capture saw a modal presentation over your app's content. 'absent' is a reading that was earned, so it is safe to act on; 'unknown' says the capture could not tell and is not 'absent'. The field is absent on other platforms. See Presented surface.

client.observability.events({ cursor, limit }) reads the session event timeline as paged JSON entries. Use nextCursor from the previous page to continue from the daemon-owned events.ndjson file without replaying already uploaded/displayed events. Cursors are absolute and survive the file's size rotation; a cursor older than the retained window rejects with COMMAND_FAILED, details.reason: "EVENT_LOG_CURSOR_EXPIRED", and details.earliestCursor to resume from. The timeline leaves out user-entered content such as typed text, selector values, and payloads; see Sessions for what it keeps.

client.observability.audio() mirrors audio probe start|status|stop. Use it to collect compact RMS/peak dBFS buckets while other session actions continue:

await client.observability.audio({
  platform: 'web',
  action: 'probe',
  probeAction: 'start',
  durationMs: 10_000,
  bucketMs: 1_000,
});
await client.interactions.click({ platform: 'web', ref: '@e4' });
const audio = await client.observability.audio({
  platform: 'web',
  action: 'probe',
  probeAction: 'status',
});
await client.observability.audio({ platform: 'web', action: 'probe', probeAction: 'stop' });

Pass platform: 'web', 'macos', 'ios' (simulators), or 'android' (emulators on macOS hosts). See Audio probes for what each platform samples, permissions, and unsupported targets.

client.observability.perf() requires an options object with an explicit area; a call without one fails with an error that names the replacement. The option shapes mirror the perf CLI forms:

• Frame health: { area: 'frames' }
• Memory: { area: 'memory', action: 'sample' }, or { area: 'memory', action: 'snapshot', kind: 'android-hprof' | 'memgraph', out } to write an artifact to disk
• CPU profile: { area: 'cpu', subject: 'profile', action: 'start' | 'stop' | 'report', kind: 'xctrace' | 'simpleperf', out }
• Trace: { area: 'trace', action: 'start' | 'stop', kind: 'perfetto' | 'xctrace', out }

To start an xctrace capture, also pass a template such as 'Time Profiler' or 'Animation Hitches'.

On Android, each successful perf({ area: 'frames' }) read and each open of an app resets the frame window, so call it, perform a transition or gesture, then call it again to inspect that window. See Performance snapshots for result fields and platform support.

client.recording.record({ action: 'start', path, quality: 'medium' }) starts a recording with medium output quality.

client.capture.screenshot({ path, scale: 0.3 }) captures a screenshot at 30% of its original width and height. scale accepts 0.01 through 1.

client.batch.run({ steps }) accepts structured steps: { command: 'open', input: { app: 'settings' } }. Step input uses the same fields as the matching client command. Daemon-shaped positionals/flags steps are internal to the daemon and not part of the client API. To run daemon-shaped batches through your own dispatcher, see Batch orchestration for custom transports.

Android installFromSource()

import { createAgentDeviceClient } from 'agent-device';

const androidClient = createAgentDeviceClient({ session: 'qa-android' });

const installed = await androidClient.apps.installFromSource({
  platform: 'android',
  retainPaths: true,
  retentionMs: 60_000,
  source: { kind: 'url', url: 'https://example.com/app.apk' },
});

await androidClient.apps.open({
  platform: 'android',
  app: installed.launchTarget,
});

console.log(installed.packageName, installed.launchTarget);

if (installed.materializationId) {
  await androidClient.materializations.release({
    materializationId: installed.materializationId,
  });
}

await androidClient.sessions.close();

On Android, a successful installFromSource() response includes the identity you need to relaunch the app:

• packageName
• launchTarget

If the daemon cannot determine the installed app's identity, the request fails instead of returning an empty success payload.

URL source rules

installFromSource() limits URL sources:

• Private and loopback hosts are blocked by default.
• URL sources from any public host may point directly to an installable, including a bare iOS .ipa, or to a .zip, .tar, .tar.gz, or .tgz archive containing exactly one.
• For existing reachable artifact URLs, use source: { kind: 'url', url: ... }.
• For local artifacts, use source: { kind: 'path', path: ... } or the CLI install/reinstall commands.
• When a remote daemon resolves CI artifacts server-side, pass a GitHub Actions artifact source:

await client.apps.installFromSource({
  platform: 'android',
  source: {
    kind: 'github-actions-artifact',
    owner: 'acme',
    repo: 'mobile',
    artifactId: 1234567890,
  },
});

Remote daemons may also support { kind: 'github-actions-artifact', owner, repo, artifactName } or { kind: 'github-actions-artifact', owner, repo, runId, artifactName }. The local client passes these payloads through unchanged; it does not authenticate with GitHub or download the artifact.

Android .apk and .aab URL sources resolve package identity from the downloaded install artifact. Archive URLs may contain one installable .apk, .aab, .ipa, or iOS .app, including inside nested archives.