Agent Browser Now Available for Telnyx AI Agents

Telnyx Agent Browser is now available on api.telnyx.com. It gives an AI agent its own isolated browser session through an API, limited to the sites you allow, so the agent can use websites that have no API.

What's new

• Isolated browser sessions: POST /v2/browser_sessions starts a short-lived Chromium session. A session lives 15 minutes by default (up to 30) and closes after 5 idle minutes by default, so nothing carries over between tasks.
• Allowed-domain control: You set policy.allowed_domains when you create a session, with 1 to 50 patterns. example.com matches only that host, and *.example.com matches its subdomains but not the apex, so list both to cover a site and its subdomains. A navigation outside the list fails with policy_denied.
• Pages read as labeled elements: The observe action returns the rendered page as a text tree with element references. An agent clicks, fills and selects by reference instead of pixel coordinates.
• Ten action types: navigate, observe, click, fill, select, press, scroll, wait_for, screenshot and batch. A batch runs 1 to 50 actions in order in a single request.
• Screenshots as artifacts: The screenshot action returns an artifact, and a signed download path retrieves the image.
• Asynchronous session creation: Creating a session returns provisioning right away. Poll GET /v2/browser_sessions/{id} until the status is ready, then send actions.
• Idempotency keys: Pass an idempotency_key on session creation or on an action so a retried request returns the original result instead of repeating it.

Why it matters

• Plenty of useful work, like booking a slot or filling a vendor form, lives on websites with no API, and an agent with only API access stops there.
• The boundaries are set before the agent starts: which sites it can reach, how long the session lives, and what it can do.
• Every action returns structured JSON, so the result goes straight back to a model as tool output.
• Failures show up on the action itself, with a code such as policy_denied or stale_observation, so an agent can tell a blocked site from an out-of-date element reference.

Example use cases

• A scheduling agent books a demo slot on a company site that has no booking API.
• An operations agent fills in a vendor signup form on behalf of a team.
• A research agent reads a pricing table on a JavaScript-rendered page that search snippets miss.

Getting started

1. Create or copy a Telnyx API key in Mission Control.
2. Create a session with the sites your agent may visit, then poll until it is ready:

POST /v2/browser_sessions
{
  "policy": { "allowed_domains": ["example.com"] }
}

3. Send navigate, then observe, to /v2/browser_sessions/{id}/actions. Use the observation_id and a ref from the observation to click, fill or select.
4. Navigation invalidates the last observation, so run observe again before the next interaction by reference.
5. Close the session with DELETE /v2/browser_sessions/{id}.

Learn more in the Agent Browser docs, the getting started guide and the Agent Browser product page.