Gateway endpoints

Call any of 947 provider endpoints with one key: search the catalogue, send the provider's own parameters, pay per call.

Verified jobs (POST /v1/execute) cover the work where a wrong answer costs you: Arettic picks the tool, checks the result and charges only if it passes. Everything else your agent needs goes through the gateway: 947 endpoints from 70 providers (840 live now), with the same key and the same balance. You send the provider's own parameters, you get the provider's own answer, and you pay per call. A call the provider fails or times out costs nothing.

1. Find an endpoint

Search the catalogue with no key. Every word must match the id, provider or name; filter by category, provider or availability. Each endpoint has an id, its method and path, the path_params it needs and its price_per_call. The whole list is also on Endpoints.

curl "https://api.arettic.com/v1/endpoints?q=google+maps"

2. Call it

POST /v1/call with the endpoint's id, plus query (for GET endpoints), body (for POST endpoints) and path_params (for each {name} in the path), exactly as the provider's API takes them. Send max_price in credits to refuse anything dearer.

curl https://api.arettic.com/v1/call \
  -H "Authorization: Bearer $ARETTIC_KEY" \
  -H "Content-Type: application/json" \
  -d '{"endpoint": "serper/google-maps-search", "body": {"q": "coffee in Bangalore"}}'

const r = await client.call({
  endpoint: "predictleads/company-job-openings",
  path_params: { company_id_or_domain: "stripe.com" },
});
console.log(r.status, r.charged, r.data);

r = client.call(endpoint="serper/google-news-search", body={"q": "AI agent funding"})
print(r["status"], r["charged"], r["data"])

{"name": "call_endpoint", "arguments": {"endpoint": "serper/google-maps-search", "body": {"q": "coffee in Bangalore"}}}

{
  "call_id": "c0a8e7a2-…",
  "endpoint": "serper/google-maps-search",
  "provider": "Serper",
  "status": "succeeded",
  "charged": { "credits": "9", "usd": "0.009" },
  "latency_ms": 840,
  "data": { "places": [ … ] }
}

What you pay

• Each endpoint has one price per call for your plan (price_per_call); Pro, Scale and Max pay less than Pay as you go. Prices per plan are on Prices.
• The price is held before the call and charged when the provider answers. If it errors, refuses or times out (60 seconds), the hold is released and refunded shows it. When the provider reports a lower cost for the call, you pay the lower price.
• Calls count towards the agent's monthly budget, like purchases. Test keys (sk_test_…) check and price a call without making it.
• There's no pass rule here: a provider's "no results" is still an answer, and charged. For checked results use the verified jobs.
• Nothing you send or get back is stored. Your dashboard lists each call (endpoint, status, charge, latency) under Activity → Endpoint calls.

On-request endpoints

Some endpoints act on resources held in a shared account (agent inboxes, phone agents, code sandboxes, browser sessions, page monitors, cloned voices), send messages or codes to people, or list everything in that account. One customer could then see or change another's, so these are listed but opened per customer: ask [email protected]. Calling one returns endpoint_on_request.

Errors

Updated 2026-10-09 · This page as Markdown · JSON