Gathering is
the default.
Give people and agents the same place to see the work - the goal, the changes, the decisions, and what’s left.
One shared workspace
Everyone sees the same goal, changes, and decisions.
Your agents stay yours
Codex, Claude Code, OpenCode - Campfire sits between them.
Work survives the session
Close the terminal. Keep the goal, the history, what's left.
Useful access, clear boundaries
Agents only get the capabilities they need here.
[ What is Campfire? ]
A shared place that runs with the agents you already have.
Give people and agents the same workspace. Start with a goal. Keep the changes, the findings, and the decisions together so the next person - or agent - does not start from an empty prompt.
Not an orchestrator
Campfire does not replace Codex, Claude Code, OpenCode, or your own agents. It sits between them so the work has a home.
Humans stay in the loop
When a decision is needed, a teammate answers in the workspace. That answer is part of the work. The next agent can see it.
| Slack threads | Shared docs | Multi-agent graphs | Agent runtimes | Campfire | |
|---|---|---|---|---|---|
| Shared goal everyone can see. The thing the team is trying to do lives in one place. | partially supported | partially supported | supported | not supported | supported |
| Survives the session. Close the terminal. The work is still there. | not supported | supported | partially supported | not supported | supported |
| Agents you already use. Codex, Claude Code, OpenCode stay your agents. | supported | supported | not supported | not supported | supported |
| Human decisions in the work. A teammate answers, and the next agent can see it. | partially supported | partially supported | not supported | not supported | supported |
| Capability boundaries. Each agent only gets what it needs here. | not supported | not supported | partially supported | supported | supported |
| History of who did what. You do not reconstruct the story from three transcripts. | not supported | partially supported | partially supported | partially supported | supported |
| Local-first. The workspace runs on your machine. | not supported | partially supported | not supported | partially supported | supported |
[ Pieces ]
The work is made of a few plain parts.
You should not have to design a multi-agent system just to get two agents working on the same problem. Create a workspace. Add the people and agents. Give them a goal. Campfire handles the shared state underneath.
- Workspace- the place the work lives
- Goal- what the team is trying to do
- Session- a stretch of agent or human work
- Decision- a human answer that changes the plan
- Capability- what an agent is allowed to do here
- History- who did something, and what it changed
[ Work that lasts ]
Close the terminal. Keep the work.
An agent session ends. Yesterday that meant the context disappeared. Campfire still has the goal, what changed, what was decided, and what’s left. The next agent does not start from an empty prompt.
See use cases[ Humans in the loop ]
Agents can do the work. People still decide.
- One shared workspace - people and agents see the same goal and status.
- Decisions recorded in the work, not scattered across DMs.
- Capability limits so agents only get what they need in this workspace.
- History you can audit: who acted, what changed, and when.
- No orchestration graph - Campfire does not replace your agents.
[ Install ]
Install Campfire
Campfire is open source and runs locally on your machine. Pick an install method, copy the command, and run it in your terminal.
Download Campfire:
curl -fsSL https://boringinfra.company/campfire/install.sh | shpress 'c' to copy
[workspace]
name = "billing-flow"
[goal]
title = "Ship usage-based billing"
[capabilities]
allow = ["repo:read", "repo:write", "tests:run"]
deny = ["deploy:prod"]
$ campfire
Welcome, Sergio.
✓ Your local profile is ready
✓ Codex configured — restart Codex to enable Campfire
✓ OpenCode configured — restart OpenCode to enable Campfire
Open your agent and say what you're working on.
It can create the workspace and goal with you.
Viewer: http://localhost:9415
Joining a team? Use your invitation file.FAQ
More detail lives in the docs and on GitHub.
Yes. Campfire is open source and in active preview on GitHub. You can read the code, run it locally, and contribute. We ship fast, but expect rough edges while the product hardens.
Campfire is not an agent orchestrator, a replacement for Codex or Claude Code, or a hosted platform that runs your agents for you. It is a workspace layer: shared context, decisions, capabilities, and history.
Run campfire to register locally, then campfire up to start the API and viewer, then campfire status to confirm everything is healthy. Onboarding a workspace is an explicit campfire onboard step when you are ready.
Harnie and Meno focus on harnessing and running agents in your environment. Campfire gives those agents (and your teammates) a shared place to see the goal, the work, and the decisions - without replacing either tool.
No. Keep using the agents you already trust. Campfire sits beside them so the output, boundaries, and decisions stay in one workspace instead of disappearing when a session ends.