Reference
Connectors
Every integration and what it is used for.
On this page
Connectors are how Jarvis works with the tools your team already uses: AI models, code hosts, ticket boards, design files, your servers and your phone. Each one is connected by an administrator, once, and then everyone whose group allows it gets the features it brings.
If a feature on this page is missing from your workspace, the connector behind it probably has not been set up. Ask your administrator.
What each connector lets you do
| Connector | What your team can do with it |
|---|---|
| Claude | Agents that write code with Claude Code. Drafted acceptance criteria, outcome reviews, worklog wording and Jarvis replies. |
| OpenAI | An alternative AI model for plans, drafts and Jarvis replies. |
| Codex | Agents that write code with Codex. |
| GitHub | Agents working on a branch per ticket, pull requests opened after approval, review comments sent back to agents, live pull request status on the ticket, and issues imported as tickets. |
| GitLab | Agents working on a branch per ticket, on GitLab.com or your own GitLab, plus importing existing clients and projects from it. |
| Jira | Tickets, comments, attachments, story points and sprints brought in from your Jira site. |
| Trello | Tickets kept in step with a Trello board, both ways. |
| Figma | Figma frames linked from a ticket turned into images an agent can work from. |
| Your servers | Approved work deployed automatically to your hosting. |
| The notifications that need someone (a run waiting for approval, a run that failed, a run asking for permission) sent to people's phones. | |
| ElevenLabs | A custom voice for Jarvis. Without it, your browser's voice is used. |
AI models
Agents and the AI features in Jarvis need at least one AI model connected. Your administrator chooses which models are available and which one Jarvis speaks through. As a user, you choose the model indirectly, by choosing the agent: each agent is set to a model. See AI coding agents.
If no AI model is connected, the workspace still works. Jarvis answers with plain summaries, and worklogs are still drafted from what happened. Drafting criteria, reviewing outcomes and agent work need a model.
Code hosts
With GitHub or GitLab connected, agents can work on your repositories. Each ticket gets its own branch, and approved work arrives as an ordinary pull request your team reviews like any other.
On GitHub, Jarvis also follows the pull request after it is opened: the ticket shows whether it is open, merged or blocked, review comments can be sent back to the agent with Address review comments, and people who can approve may be offered a merge button.
Clients and projects can be imported from either host, so you do not have to type in what already exists.
Ticket boards
With Jira connected, a project can be pinned to a Jira project. Its tickets come into Jarvis with their comments, attachments and story points, and each ticket exists only once even if two projects point at the same board. Sprints are read from Jira and shown on the Sprints screen. A ticket Jira never sized keeps any estimate made in Jarvis.
With Trello connected, a project's tickets stay in step with a Trello board.
GitHub issues can be imported into the project for that repository.
Design files
With Figma connected, linking a Figma frame in a ticket lets an agent working from a design see it as an image. A link to a whole file, rather than one frame, is mentioned to the agent but not turned into an image.
Deploying
When your project is set up to deploy, approving agent work also sends it to your hosting. Nothing is ever deployed without an approval. If a project is not set up to deploy, approved work stops at the pull request and is marked Completed. See Runs and the approval gate.
Notifications
The bell in the workspace shows what needs you. With WhatsApp connected, the important ones also reach your phone, with a link back to the right screen. Messages go to the contact number on your account, so keep it up to date.
Workspace-wide or per project
A connector can serve the whole workspace or a single project. That is how one client's projects can use a different code host account or deploy target from everyone else's. Your administrator decides.
Keys stay with administrators
Connecting a tool means giving Jarvis a key for it. Once saved, a key is never shown again on screen, is never given to an agent, and never appears in a run's output. See How Jarvis keeps work safe.
Something missing or unclear? Tell us.