Track

Track is one of WebAgent's typed APIs. You describe a monitoring intent; it repeatedly fetches data from web pages on a schedule, compares against a baseline, and notifies you through your chosen channel when a change hits.

Unlike DoAnything / DeepResearch / WebSearch: those "run once, produce a result". Track's resource is a monitor — a long-lived object that runs itself repeatedly on a schedule, producing one run per tick.

When to use it

• What you care about is how something changes over time, not a one-off result right now.
• You want to be pushed a notification when a change happens, rather than polling yourself.
• The monitoring can stay running long-term — a monitor exists until you delete it.

When you only want a one-off result, use WebSearch or DoAnything.

Create a monitor

HTTP endpoint:

POST /v1/projects/{pid}/track/monitors
Authorization: Bearer wa_...

Request fields:

Field	Type	Required	Description
intent	string	Yes	The monitoring intent, natural language
notify_channel	object	Yes	The notification channel — see notify_channel below
schedule	object	No	The trigger schedule — see schedule below. Omit = use the default schedule
target_urls	string[]	No	Restrict which pages to monitor; omit and the agent finds them itself
extraction_schema	object	No	Specifies which structured fields to extract from the page
trigger_dsl	object	No	The trigger condition — only when met does it count as a "change hit"
stop_condition_dsl	object	No	The stop condition — once met, the monitor stops automatically
profile_id	string	No	Reuse the login state of a Profile

For the full schema, see CreateMonitorRequest in the OpenAPI spec. schedule and notify_channel are nested objects, structured as described in the two sections below.

schedule

schedule.kind decides when the monitor runs:

kind	Description
interval	Fixed interval; pair with interval_seconds
cron	Cron expression; pair with cron
event	Triggered by an external event; pair with event_filter
autonomous	The agent decides itself when the next run should be

notify_channel

notify_channel.kind decides where a change hit is pushed:

kind	Description
callback_url	POSTs to your URL; pair with url
global_webhook	Reuses a project-level webhook; pair with webhook_id
console_inbox	Pushes to the Console inbox, suited for a human to read

Example

Create a monitor with intent + schedule + notify_channel together:

from web_agent.v1 import Client

async with Client(api_key="wa_...", project_id="proj_demo") as client:
    monitor = await client.track.create(
        intent="Notify me when Apple's stock price drops below $200",
        schedule={"kind": "interval", "interval_seconds": 3600},
        notify_channel={"kind": "callback_url", "url": "https://hooks.example.com/track"},
    )
    print(monitor["id"], monitor["status"])

Baseline and change detection

A monitor's first run establishes a baseline (baseline_extracted). Every run afterward is compared against the baseline, and a notification is sent only when trigger_dsl is met. last_tick_at / last_tick_n record the time and sequence number of the most recent run, and consecutive_failures counts consecutive failures — too many consecutive failures and the monitor enters an error state.

Lifecycle

Once created, a monitor exists long-term until you delete it. Common operations:

Operation	Description
pause / resume	Pause / resume scheduling
run_now	Run once immediately, without waiting for the schedule
cancel / delete	Stop / delete the monitor
patch	Change the intent, schedule, channel, etc.
refine	Adjust the monitoring intent in natural language
list_runs / get_run	Query historical runs
list_deliveries / retry_delivery	Query notification delivery records; failed ones can be redelivered
intervene / message	Respond when the monitor hits a login, captcha, or other situation needing human intervention

The event stream endpoint …/track/monitors/{mid}/events follows the SSE conventions; follow each tick of the monitor in real time.

Next steps

• Profiles — reuse login state when monitoring sites that require a login
• WebSearch — use this when you only want a one-off result
• API Overview — shared conventions; the full field set is in the OpenAPI spec