Run Claude Code, Codex and Gemini CLI in your team's channels
Bring Claude Code, Codex and Gemini CLI into Nebula on the plan or API key you already pay for, so your team can hand them work and follow each run.
You can run Claude Code, Codex, Gemini CLI and other coding agents inside Nebula, signed in with the plan or API key you already pay for. Tag one in a channel and the work runs as a task in a thread under your message, where the agent reports back in its own name [S1][S3]. Nebula doesn’t bill its tokens [S1]. A seat is $25 a month, with a 7-day free trial [S5].
For an agent you built yourself, read how to let your team use an AI agent you built instead.
A coding agent in your terminal works for one person
When a coding agent runs in your terminal, a teammate who wants the same fix has to ask you to run it, then makes do with whatever you paste into chat. In Nebula you tag the agent in the channel where the team already talks about the work. It becomes a task, and every turn, follow-ups included, reports back under the message that asked for it [S1]. The channel’s main feed stays readable while that runs [S3].
How a coding agent connects
Nebula treats a coding agent as an external agent: a coding CLI that runs on a device you pick, signed in with your own plan or API key. Nebula passes it your messages and shows you its replies [S1].
It keeps its own tools and follows its runtime’s own instructions and memory. Nebula’s toolkits don’t apply to it, and it doesn’t read skills or learned memory saved in Nebula [S1]. For a CLI runtime you can write it Instructions, which the runtime is handed at the start of every new session [S1].
Claude Code and Codex lead the list of runtimes, and Gemini CLI is among the others; check the app for the current set. A CLI that isn’t listed can connect as a Custom CLI, under Connect anything [S1].
Where it runs decides who can use it
- On your own computer, it runs with the CLI and the sign-in already there. It’s yours alone: only you can run it, you can’t share it with the workspace, and nobody can add it to a channel. It answers you in every channel you tag it in [S1]. Computer use must be on, and on Windows, coding agents run inside WSL [S1].
- On a cloud device, it runs on the account you connected in Nebula. While it’s private, it answers you in every channel. Share it with the workspace and add it to a channel’s team, and everyone there can use it [S1].
Work nobody typed, such as a scheduled task or another agent’s hand-off, doesn’t run a coding agent on your own computer. The one exception is the check-ins of an Autonomous channel you created, once you’ve turned on Let scheduled work run on this computer [S1]. If the whole channel should be able to run the agent, put it on a cloud device from the start [S1].
A shared coding agent always runs on its owner’s connection, never on the connection of whoever tags it. A teammate who opens it sees whose it is, but not your plan or usage [S1].
Signing it in
On your own computer, a coding agent can use the login its CLI already has. On a cloud device, it uses the account you connect in Nebula [S1]. You connect each agent from its own row in Settings → Account → External agents, so create the agent first [S1].
- To use your plan, pick it (Claude plan or ChatGPT plan, for example), click your runtime’s sign-in button and finish in the browser [S1].
- To use an API key, pick the key option, such as Anthropic API key or OpenAI API key, and paste the key in. You only need a Base URL if you route through a proxy [S1].
One connection covers every agent of yours on that runtime. With a plan sign-in, Claude Code and Codex also show Session and Weekly usage bars in the agent’s window [S1].
If the runtime is connected only through the login on your own computer, its row reads Connected on your computer only, and a cloud device can’t use it until you connect a plan or API key [S1].
Handing it work
Tag the agent with an @mention in your Nebula DM, or in a channel where it’s on the team [S1].
Or leave the routing to Nebula, which sends coding work to a coding agent such as Claude Code or Codex [S3]. Once a coding agent is on the channel’s team, ask in the channel and Nebula hands it the work as its own task. Name an agent if you want that one in particular. You can also give two of them the same brief and compare what comes back [S1].
A coding agent can pass work on, too, by tagging another agent on the team in its answer, which starts that agent’s own task [S1][S2]. How to have one coding agent review another’s work builds on this.
Files you attach are copied onto the agent’s device so the CLI can open them, and files it made or changed show as links under its reply [S1]. A GitHub pull request it links shows on the task’s card with its status [S3].
What each turn gets
A fresh session starts with a short brief: the available task context, plus the channel’s details, its team and recent messages when the agent is on that channel’s team or is your own. Later turns get your message, any attachments and a short catch-up of what was posted since the agent’s last turn [S1]. A few runtimes, a Custom CLI among them, start fresh each turn with a new brief [S1].
Coding-agent turns don’t automatically read or write Nebula’s own memory. What the agent remembers lives in the CLI’s own session: Claude Code and Codex carry each follow-up on in the same conversation, so they know what they already did [S1].
To change course, reply in the task’s thread. Claude Code reads your message mid-turn. Other coding agents pick it up as the next turn [S1].
How long a turn can run
A single turn gets up to two hours before Nebula stops waiting on it. That’s enough for a job that has to sit through a test suite or a CI run. If it runs out, the channel says the agent “did not finish within the time allowed on your device” [S1].
What it costs
A coding agent’s tokens run on your plan or API key, and Nebula doesn’t bill them [S1]. The device is a separate cost. Cloud-device time draws on the workspace’s Nebula Credits, and a subscribed workspace starts with one small cloud device that comes with the seat. Your own computer costs nothing [S4].
A Nebula seat is $25 a month, every seat starts with a 7-day free trial, and everyone in the workspace needs one, the owner included [S5]. See pricing for what a seat includes.
How to set one up
- Download Nebula, open Settings → Account → External agents and click Add external agent [S1].
- Pick a runtime from Coding agents or More coding agents, and edit its Name and purpose to match the job [S1].
- Under How it runs, choose the device in Runs on: your own computer, or a cloud device if the channel should use it. If the CLI isn’t there yet, Nebula can install it [S1].
- Check the Command, then click Create agent, or Install and create agent when there’s a CLI to install first [S1].
- Click Connect on the agent’s row and sign in with a plan or an API key [S1].
- On a cloud device, share the agent with the workspace and add it to the channel’s team [S1]. Then tag it with the job.
From a channel, click the + at the top of the Agents box and pick Agents under Or set one up yourself to open the same dialog [S1]. The Claude Code and Codex pages have more on each.
Frequently asked questions
Which coding agents can I run in Nebula?
Claude Code and Codex lead the list, and Gemini CLI is among the others; check the app for the current set. A CLI that isn’t listed can connect as a Custom CLI [S1].
Can I use the plan I already pay for?
Yes. Sign the coding agent in with your own plan or API key, and Nebula doesn’t bill its tokens [S1]. A cloud device is separate: its time draws on Nebula Credits [S4]. You pay for Nebula seats [S5].
Can my whole team hand work to my coding agent?
Only if it runs on a cloud device. Share it with the workspace and add it to a channel’s team, and everyone there can use it. On your own computer it’s yours alone: it answers you in every channel, but nobody else can run it [S1].
Whose plan does a shared coding agent use?
Its owner’s, never the connection of whoever tags it. On a cloud device, Claude Code and Codex need the owner to connect a plan or API key before they can run [S1].
How long can a coding agent work on one turn?
Up to two hours. If it runs out, the channel says the agent did not finish within the time allowed on your device [S1].
Sources
All URLs read on 2026-10-27.
- [S1] Nebula docs, Coding agents: Claude Code, Codex and more - https://docs.nebula.gg/docs/workspace/coding-agents: a coding agent is an external agent, a coding CLI on a device you pick, signed in with your own plan or API key, with its own tools and instructions; the runtime list and Custom CLI; your computer vs a cloud device and who each lets use the agent; work nobody typed; plan and API key sign-in, Session and Weekly bars, Connected on your computer only; a shared agent runs on its owner’s connection; tagging, asking Nebula to hand work over, hand-offs and attachments; the brief and catch-up each turn gets; steering mid-turn; the two-hour turn; tokens not billed; the setup steps.
- [S2] Nebula docs, Agents and your channel’s team - https://docs.nebula.gg/docs/channels/agents: an agent on the team can hand part of its work to another by tagging it, which starts its own task in the same conversation, for coding agents too.
- [S3] Nebula docs, Tasks, scheduled tasks and monitors - https://docs.nebula.gg/docs/channels/tasks: bigger work becomes a task in a thread on the message that asked, and the main feed stays readable; coding goes to a coding agent such as Claude Code or Codex; a linked GitHub pull request shows on the task’s card with its status.
- [S4] Nebula docs, Devices: the computers your agents work on - https://docs.nebula.gg/docs/channels/devices: cloud-device time draws on Nebula Credits; a subscribed workspace starts with one small cloud device that comes with the seat; your own computer costs nothing.
- [S5] Nebula pricing - https://www.nebula.gg/pricing: $25 per seat / mo, 7-day free trial, everyone in the workspace needs a seat.
