Connectors
How connectors work, and when the browser is used instead
A connector performs a structured operation against a service API through the backend, while a browser action works through the website interface; resolvenengine chooses the connector when your request is about your own data in a connected service, and the browser when the task is about the page in front of you or open research.
Last updated 18 September 2026
The routing order
Routing is decided by rules before the model starts working, so the same request behaves the same way every time.
- 1An explicit instruction wins. "Use Notion" or "do this in the current tab" overrides everything else.
- 2Page work stays in the browser. Anything about the tab you are on is handled there.
- 3Open research stays in the browser, because the answer is not in your own accounts.
- 4Your own data goes to a connector: possessive and locative phrasing such as "my inbox" or "in my workspace".
- 5A weak signal, such as only naming a product, does not select a connector on its own.
- 6Several plausible candidates produce a clarifying question rather than a guess.
- 7A disconnected service is stated out loud instead of silently falling back to the browser.
Why a connector is preferred for your own data
- It uses the provider's own permission model, so what you granted is exactly what can run.
- It returns structured fields, including real status names and timestamps, instead of scraped text.
- It is auditable: each step reports the provider and operation that produced the evidence.
- It is faster, because it skips page loads entirely.
Released connectors
| Service | Reads | Writes |
|---|---|---|
| Gmail | Search and read messages | Not offered |
| Google Drive | Search files, read supported Docs | Create a folder |
| Google Calendar | Calendars, events, details | Create an event |
| Notion | Search and read pages, databases | Create page, append text, comment |
| Slack | Channels, messages, threads, members | Post message, create or join a channel |
| ClickUp | Search tasks, workspace structure | Create a task |
| GitHub | Repos, files, issues, pull requests | Repos, commits, issues, pull requests |
| Google Analytics 4 | Properties, traffic, realtime | Read-only |
| Search Console | Performance, sitemaps, URL status | Read-only |
Every write is an explicit request and follows the approval policy you set for that operation.
Frequently asked questions
Why did it use the browser when I named a tool?
Naming a product alone is a weak signal, it often means research about that product. Say "in my Notion" or "my Slack channel" to point at your own data.
Can I add my own service?
Yes. Custom HTTPS/OpenAPI connectors and remote MCP servers can be added from the Connectors panel.
Related guides
Connect Gmail, Drive and Calendar safelyWhy Google sign-in and Gmail consent are separate steps, which permissions each Google connector requests, and how to revoke access.Add a custom connector or a remote MCP serverAdd your own HTTPS/OpenAPI service or a remote MCP server as a connector, with owner-bound credential storage, tool discovery and approval policies.Allow, Ask, Deny: the permission model explainedSet a policy for every single operation, require approval for writes, and understand why a denial is never bypassed.
Try it in your own Chrome
Pro is $19 USD per month and uses your own AI API key, so the provider bills model usage at their own rate.