Jotform AI Agent Chat
This integration sends each customer message to a published Jotform AI Agent and returns the agent's next reply. It keeps one Jotform chat and its message history for the sandbox session, so later turns retain the conversation context.
The integration reproduces the transport used by the public agent page. It does not sign in to Jotform, read your form submissions, or change the agent configuration.
Required secrets
| Secret | What it is |
|---|---|
JOTFORM_AGENT_ID | Agent identifier from the direct Jotform AI Agent URL. |
Create the agent in Jotform
- In your Jotform workspace, create an AI Agent, or start from one of Jotform's agent templates.
- Give the agent its instructions and any knowledge it needs to answer.
- Publish the agent so it answers on its own direct link.
Jotform asks you to verify your account email before it lets you create new assets, so confirm that first if the workspace blocks you.
Get the agent ID
- In Jotform, open the agent and select Publish.
- Copy the direct link. It looks like
https://agent.jotform.com/0198c0e9a1b2c3d4e5f6. - Use the final path segment as
JOTFORM_AGENT_ID.
Send a test message on that page before you connect the agent to Spec27. An agent that is unpublished or restricted by domain will reject the adapter's requests.
Set it up in Spec27
- Open your project and select Agents.
- Choose Jotform AI Agent Chat from the registry.
- Enter
JOTFORM_AGENT_IDas a project secret. - Run the example input from the agent page to confirm the agent replies.
Turn support
This integration supports single-turn and multi-turn evaluations. It creates the Jotform chat on the first turn, then replays the message history Jotform returns on every later turn.
Limitations and notes
- The integration returns the final assistant message of each streamed response. It does not serialize cards, forms, file uploads, or voice replies.
- It allows 15 seconds for chat creation and 120 seconds for each message, which covers the hosted actions a Jotform agent can run mid-reply.
- A response that carries no assistant message fails the turn rather than returning empty text.
- Evaluations create real conversations and can trigger any action configured on the agent. Use a test agent when the actions can change customer or business data.