Connector Studio

• Build a connector by chat. Name the API, describe the calls you need, and the agent writes them: path, query and header parameters, and for write calls a JSON body.
• Test before switching on. Every call can be tried on the live API from the draft. You see the real response before anything reaches your team.
• Start from an OpenAPI file. Give the agent the address of an OpenAPI 3 (JSON) description and it suggests calls from it. You choose which ones to add.
• Read and write. GET calls run straight away. Every POST, PUT, PATCH or DELETE call stops and asks you first, showing the address and the data it will send, and runs once after you approve.
• Versions and rollback. Each time you switch a connector on, a version is frozen. If a change breaks something, the agent can return to an earlier version.
• Install on other agents. An active connector can be added to any agent in your company.

1. Install it. Open the Skill Marketplace, find Connector Studio, and install it on the agent that will build connectors.

2. Describe the API. Talk to the agent in normal language:

User: I want to read exchange rates from the Frankfurter API. Base address is https://api.frankfurter.dev/v1.
Agent: Drafted a connector with one call, latest_rates (GET /latest, parameters: base, symbols). Shall I test it with base=EUR?

User: Yes, test it.
Agent: It works: the API returned today's euro rates for 30 currencies. Switch it on for this agent?

User: Switch it on, and add it to Julia too.
Agent: Done. "Frankfurter Rates" v1 is active on both agents.

3. Add the key, if the API needs one. When a connector uses an API key, bearer token, username and password, or an OAuth2 client ID and secret, the agent gives you a link to the skill's settings in the panel. Enter the key there, not in the chat.

• Keys never pass through the chat. They are entered in the panel, stored encrypted, and added to the request by the gateway. The agent never sees them, and if an API echoes a key back in a response, it is removed before the agent reads it.
• Every write asks first. POST, PUT, PATCH and DELETE calls never run on the agent's say-so. You get an approval request showing the address and the data that will be sent. Approve it and it runs once; reject it and nothing is sent.
• Private to your company. Connectors are visible only to your company's agents. They are not listed in the marketplace.

• Public HTTPS APIs only. The address must be https on port 443 and resolve to the public internet. Private networks, local addresses and IP literals are refused, for your safety and ours.
• JSON in, JSON out. Calls send JSON bodies and read JSON (or plain text) answers. Long lists are cut to 50 items, and an answer over 1 MB is refused.
• Sign-in methods. No sign-in, bearer token, API key (header or query), basic auth and OAuth2 client credentials (a client ID and secret exchanged for a token). Sign-in flows that open a login page are not supported.
• Up to 12 calls per connector. For a bigger API, build more than one connector.
• Redirects are not followed. If an API answers with a redirect, use the final address as the base.