Watches

A watch is a standing query Vaaya keeps running for you. You describe what to look for in plain language, Vaaya polls on a schedule, and only genuinely new findings surface. Creating, listing, pausing and deleting a watch are free; the scheduled runs spend from your balance under a daily budget you control.

The model

Every watch is a *worker*: a saved query plus a kind, a cadence, and a findings list. The kind names what it does and picks the machinery behind it.

Kind	What it watches
`signal`	Buying signals against a plain-English ICP: funding, hiring, launches, leadership changes, press. Polled roughly every 6 hours.
`job_search`	Roles matching a description, as they are posted.
`research`	A standing deep-research task, run asynchronously.
`custom`	Free-form: a page, a topic, a competitor, anything you can describe. The default.

Findings are de-duplicated across runs, so a watch reports what changed rather than everything it saw. They land on the Watches dashboard in the app, and are readable from any agent with worker_findings.

The tools

Watches are driven over MCP. There is no /api/run equivalent today, so an HTTP-only integration cannot create them.

Tool	What it does	Price
`worker_create`	Create a watch. Pass the query, and optionally `kind` and cadence.	Free
`worker_list`	Your watches with kind, status, cadence, last run and finding counts.	Free
`worker_findings`	Recent findings, newest first. Optional `worker_id` and `limit` (default 50).	Free
`worker_pause`	Stop a watch running, keeping it and its findings.	Free
`worker_resume`	Resume a paused watch.	Free
`worker_run_now`	Run every active watch immediately instead of waiting for the next tick.	Spends from the daily budget
`worker_delete`	Delete a watch and its findings.	Free

// "Watch seed-stage B2B SaaS in Europe that just raised."
worker_create({
  query: "seed-stage B2B SaaS companies in Europe that just raised",
  kind: "signal",
  signal_types: ["funding", "hiring"]
})
// → { ok: true, worker_id: "wrk_4b1e", kind: "signal" }

What it costs

Creating, listing, pausing, resuming and deleting a watch are free.
Each scheduled run spends from your balance at the price of whatever it calls (search, scraping, research), the same per-call prices as anywhere else in the catalog.
Spend is bounded by a daily budget, $1/day by default, which you can change. A watch stops running for the day once its budget is used.
worker_run_now spends against the same budget, so a manual run is not free.

Where findings land

Findings appear on the Watches dashboard in the app, where a signal finding can be worked into outreach directly. Agents read the same list with worker_findings, so a watch created in chat is visible to Claude Code, and vice versa.