Run multi-source deep research. The pipeline searches the web, scrapes the most relevant citations, and uses an LLM to structure the combined data into a unified answer. Async — typically 1–5 minutes. Use this when one URL or a flat search result will not answer the question (comparative analysis, multi-jurisdictional research, market intelligence). Returns a summary plus structured_data conforming to the inferred or supplied schema.

Ask multiple AI answer engines (ChatGPT, Gemini, Google AI Overview) the same question and compare their answers. Returns one result per engine — status, an answer summary, latency, credits used, and a consensus/outlier verdict — plus an AI-generated synthesis of where the engines agree and diverge. Async; typically completes within 1–2 minutes and this tool polls to completion. Use for brand/AI-SEO visibility checks ("what do AI engines say about X"), answer comparison, and geo-specific AI answers (set country). Billed per source at that Wire action's rate; failed sources are free. Set include_full_content=true only when you need each engine's raw full answer — it is large.

List the AI answer engines available to ai_visibility_search — each with its slug (what you pass as sources) and display label. Call this when you need to query a subset of engines or check what is currently enabled.

Run a natural-language task in a real cloud browser driven by an AI agent: it navigates, clicks, types, scrolls, and extracts on your behalf ("find the cheapest 65-inch TV on this site and list its specs", "fill the contact form with …"). Use when scrape cannot do the job (multi-step flows, interactions, complex navigation) and no Wire action covers the site (check wire_discover first — Wire actions are faster and cheaper). Async; runs up to ~5 minutes and this tool polls to completion. For login-protected tasks pass session_id from session_list — never put passwords in the prompt. Supply output_schema to get structured JSON back. It does not execute payments or transfer funds; such tasks are refused. Returns the task result plus run metadata (steps taken, duration, run_id).

Bulk-fetch markdown across a site. Use this when an agent needs the contents of many pages at once (catalog ingestion, site-wide RAG corpus). Pair with includePatterns / excludePatterns to scope which URLs are fetched. Returns an array of pages each with markdown and per-page status.

Discover all reachable URLs under a given site. Useful for understanding a domain's structure before crawling, or finding the sub-pages an agent should scrape. Returns lists of internal links, external links, and counts. Honors depth and limit parameters.

Get the detected changes for a monitor — each entry records when the watched content differed from the previous check, with a diff/summary (and the AI change summary when aiMode is on). Use monitor_list first to find the monitor id.

Control an existing website monitor: "pause" stops scheduled checks, "resume" restarts them (may hit the plan's active-monitor cap), "run_now" triggers an immediate out-of-schedule check (billed like a normal check), and "delete" permanently removes the monitor and its history. Use monitor_list to find the id.

Create a scheduled website monitor that checks a URL every intervalMinutes (min 15) and records a change when the content differs — optionally alerting a webhook or email. scope "page" (default) watches one URL; "site" crawls the site each run and tracks pages added/removed/changed; "wire" runs a Wire action each check and diffs its JSON. watchMode "full_page" (2 credits/check) compares the whole page; "specific_data" (3 credits/check) extracts only the fields in outputSchema with AI — ideal for price/stock/status tracking. aiMode (+1 credit/check) filters out trivial noise and summarizes real changes. Active-monitor caps per plan: Free 5, Pro 20, Scale 100.

List your website monitors, or pass id to fetch one monitor's full configuration and status (next/last check time, active state, per-check credit cost, alert settings). Use this to find a monitor's id before monitor_changes or monitor_control.

Fetch a single URL and return clean markdown by default. Set generateJson=true to also extract structured data with AI. Set useBrowser=true for SPAs and JS-heavy sites (slower and more expensive — only when needed). Returns markdown unless generateJson is true, in which case it returns the structured JSON.

Run an AI web search and return result URLs, titles, and snippets. Synchronous — returns immediately, no polling. Use this when the agent needs to discover pages relevant to a query before scraping. Returns a results array with url/title/snippet/date for each hit.

Permanently delete a saved browser session and its encrypted login data. Irreversible — the user must log in again through the dashboard to recreate it, and any monitors or requests referencing this sessionId will lose authenticated access. Find ids with session_list.

List your saved browser sessions — encrypted login states captured via the Anakin dashboard or Browser API. Each session's id is what you pass as sessionId to scrape/crawl, monitor_create, or browser_task to work with login-protected pages. Optionally filter by the website domain the session belongs to. If no session exists for a site, the user must create one interactively in the dashboard (log in once; 2FA/captchas included) — that flow cannot run from here.

Request brand-new Wire actions for a website that isn't in the catalog yet — a full catalog build. Describe the site (website_url) and what to build (goal); optionally list the discrete capabilities as actions (each becomes its own action), pin the proxy exit country, and attach a login credential to build actions behind a sign-in. Wire generates and auto-tests the scrapers, then publishes them. Asynchronous — the response's build_request carries an id and status "pending"; when it completes, its skipped list names anything the build could not deliver. Charges credits (login builds cost significantly more than public builds), refunded automatically if the build fails. Track progress with wire_build_status. Only use this after wire_discover / wire_catalog confirm no existing action covers the site.

Check on Wire builds started with wire_build. Pass the id from wire_build's build_request to get that build's full detail: its status ("pending"/"processing" → still running, poll again after ~30s), the published actions (each with the action_id to run it via wire_read_action / wire_write_action), the skipped list (capabilities the build could not deliver — always check it, delivery may be partial), and the catalog_slug once an action has published. Omit id to list your recent build requests instead (optionally filtered by status). Read-only; spends no credits.

Browse the Wire catalog. With no arguments, lists every supported website and its action count. Pass a catalog slug (e.g. "walmart", "amazon", "linkedin") to get that site's full action list with exact parameter schemas, each action's type (read/write), auth mode (none/optional/required), and credit cost — plus the login fields for credentials-mode sites. Use this to see everything a specific site can do (e.g. which read actions exist for fetching category products) before running one with wire_read_action or wire_write_action.

Find Wire actions for a task from a natural-language intent. Wire is a catalog of pre-built automation actions across hundreds of websites (Amazon, Walmart, LinkedIn, Airbnb, Zillow, and others). Actions are of two kinds: READ actions that extract data (search listings, fetch a category's products, get a product's price/specs/reviews, read a profile, pull dashboard metrics) and WRITE actions that perform interactions (log in, submit a form). Many read actions need no authentication. Applicable when a task may be achievable on a specific known site, for data extraction as well as interactions. Returns ranked candidate actions, each with its action_id, type ("read" or "write"), required/optional params, credit cost, and whether auth is needed. Run a returned action with wire_read_action (when its type is "read") or wire_write_action (when its type is "write").

List your saved Wire identities and their credentials. An identity is a named account on a site; each credential's id is the credential_id you pass to wire_read_action / wire_write_action to run actions whose auth_mode is "required". Optionally filter by catalog_id. Use this to find an existing credential before running an auth-required action (and check its status is "active", not "expired").

Sign in to a credentials-mode site and get a credential_id usable immediately with wire_read_action / wire_write_action. Provide the catalog slug and login params (the fields that catalog's login schema defines, e.g. email/password — see wire_catalog's login_input_schema). The password is never stored, only the encrypted session. Only needed for actions whose auth_mode is "required", and only for catalogs that support password sign-in; cookie-based sites use the dashboard connect flow instead.

Run a Wire READ action — one whose type is "read" (it EXTRACTS data and does not change state on the target site): search listings, fetch a category's products, get a product's price/specs/reviews, read a profile, pull dashboard metrics. Discover action_ids first with wire_discover or wire_catalog and confirm the action's type is "read"; params must match that action's parameter schema. This tool transparently polls the async job to completion and returns the extracted data. Most read actions need no auth; if the action's auth_mode is "required" (e.g. reading data behind a login), pass a credential_id from wire_identities or wire_login. For state-changing actions (type "write") use wire_write_action instead.

Run a Wire WRITE action — one whose type is "write" (it performs a state-changing interaction on the target site): submit a form, add an item to a cart, post or send content, update account settings. Discover action_ids first with wire_discover or wire_catalog and confirm the action's type is "write"; params must match that action's parameter schema. Most write actions need auth — pass a credential_id from wire_identities or wire_login. This tool transparently polls the async job to completion and returns its result. It does not execute payments or transfer funds; such actions are refused. For data extraction that does not change state (type "read") use wire_read_action instead.