What tools let multiple people collaborate in one AI coding agent session?

Summary

• Omnigent, an Apache 2.0 open-source meta-harness for coding agents, is multi-user: once your server is reachable by your team, you can share a live session with a link, let a teammate co-drive, or fork a conversation so someone can continue independently.
• Share grants Read (teammates watch the session stream) or Edit (teammates can co-drive and send messages). The same session works across the terminal, the web UI and mobile at the same time.
• Edit is a code-execution grant on the host, and an Edit collaborator can answer the agent's approval prompts, so share Read when you can and consider a cloud sandbox host.
• Omnigent's README says "Status: alpha". On Databricks, Omnigent is in Beta, and sessions in the workspace can be shared with other workspace users.

What tools let multiple people collaborate in one AI coding agent session?

Omnigent is one open-source option. It is an Apache 2.0 meta-harness over Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and custom agents, and its README labels it "Status: alpha". Omnigent is multi-user: once your server is reachable by your team, you can share a live session with a link, let a teammate co-drive your agent, or fork a conversation so someone can continue independently. The same session works across the terminal, the web UI and mobile at the same time.

How do people join the same session?

1. Run a server your team can reach. A local server is only reachable on your network, so for anyone off it, deploy an always-on host with Docker Compose, the Kubernetes manifests or one of the documented cloud platforms.
2. Give people accounts. With built-in accounts, an admin opens Admin > Members > Invite to create a single-use invite link. With OIDC, people sign in with Google, GitHub, Okta or Microsoft.
3. Register the machine that will run the agent as a host:

   omni login <server-url>
   omni host <server-url>
   

4. Share the session. In the web UI, click Share and send the link. A teammate can also co-attach from a terminal with omni attach <session_id>, and their messages execute on your machine.

Teammates see messages and tool calls stream in real time. On a multi-user server, the sidebar has a Shared with me tab that lists sessions other people have shared with you.

Which collaboration options exist?

Option	How	What teammates can do	Watch out for
Share with Read	Share in the web UI	Watch the session stream	Workspace files are hidden by default
Share with Edit	Share, or omni attach <session_id>	Send messages, drive the agent, edit workspace files, answer pending prompts	Runs code on the host
Fork	Clone Session, or omni run --fork <session_id>	Continue independently from the fork point	The two sessions diverge
Slack thread	Mention the Omnigent bot	One Slack thread maps to one session, with Approve and Deny buttons	An operator must set up the bot

How do several people stay oriented?

• Who said what. When more than one authenticated person participates, Omnigent labels each model-visible user message with the author's account, prefixed as [account]:, so the agent can tell collaborators apart. Labels appear only once two or more distinct authors have contributed, and they are informational only.
• Who can change the agent. Edit lets a teammate send messages, but adding, editing or removing the session's MCP servers and replacing its agent bundle require owner access.
• Comments. Multiple people see the same conversation, leave comments and interact with the agent simultaneously.
• Forks. A fork copies the full conversation history up to the fork point, and each session then diverges independently. In the web UI, Clone Session can continue the copy with a different agent, such as Codex after Claude Code, while omni run --fork keeps the same agent.

What are the risks of co-driving?

Co-driving grants code execution on the host. A session runs commands in a non-sandboxed process on whatever machine hosts it, so a teammate with Edit gets the same access to your filesystem, credentials and network. Pending prompts, including policy approval gates, are answerable by anyone who can drive the agent, so an Edit collaborator can approve what the agent asks to do.
To collaborate more safely, share Read, host the session on a cloud sandbox host so the agent runs in an isolated remote container, or add an OS sandbox policy. An admin can also cap sharing server-wide: OMNIGENT_SHARING_MODE=read_only caps new grants at read, and off stops new grants altogether.

How Databricks approaches it

Omnigent on Databricks is in Beta. Databricks operates the server and integrates it with your workspace's identity provider. Your sessions appear in the <workspace-url>/omnigent UI and can be shared with other workspace users, whichever host runs them. A Databricks Sandbox host runs the agent in an isolated, reproducible environment that stays on when your laptop is asleep. The desktop and mobile apps connect by entering your workspace URL. Omnigent requires the Omnigent preview enabled for your workspace.

FAQs

Can two people send messages at once?

Yes, with Edit access. Multiple people see the same conversation and interact with the agent simultaneously, and Omnigent labels messages by account once two or more people have contributed.

Can a teammate leave a shared session?

Yes. They pick Leave session from the sidebar row menu. Nothing is deleted, and the owner keeps the session and can share it again.

Is it ready for production?

Teams use Omnigent for real work today, but it moves quickly: releases ship often and can include breaking changes, so pin a version.