TopolearnWebhooks connect agents to tools that have no built-in trigger, like your own website or internal systems.
A webhook is an automatic message one system sends to a web address when something happens: "new signup", "order paid". A webhook trigger gives your agent such an address.
When the other system sends data to the URL, usually as JSON (a text format that writes data as "name": value pairs, like the example), Gumloop accepts it, replies right away with a success message, and runs the agent in the background. The agent's answer is not sent back to the caller; have the agent post its result somewhere (Slack, a CRM, an email).
Getting the data into the prompt:
- Insert the Raw JSON badge into your prompt: "A new signup arrived: [Raw JSON]. Research the company and post a summary in #sales."
- Or turn on Send raw JSON instead of a prompt and let the agent's instructions explain what to do.
Protect the URL: it contains a secret part and works as the password itself, so anyone who has it can start your agent. Don't paste it into public places. Each trigger accepts a limited number of calls per minute; extra calls are refused with a "try again later" response.
For developers, the Gumloop API goes further: create and update agents, start sessions (conversations) and send messages, and even call agents through an OpenAI-compatible chat interface. It uses a Gumloop API key from your profile settings, with Python and JavaScript SDKs (ready-made code libraries) and a command-line tool. Non-programmers can ask a developer to wire this up, or use webhooks for the simple cases.
Example
{
"event": "signup",
"customer": {
"email": "ada@example.com",
"company": "Example Labs",
"plan": "trial"
}
}Remember
- A webhook trigger gives the agent a URL any system can call with data
- The caller only gets an acknowledgement; the agent must post its result itself
- The URL is a secret; the API offers full programmatic control
Watch out for
- Expecting the webhook call to return the agent's answer, like a normal web form.