Prasenjit Sarkar (@stretchcloud) on X

What it claims

Prasenjit Sarkar (@stretchcloud), who built Campfire, presents it as a free, open-source, self-hosted alternative to paid multi-agent coding orchestrators. The post was written in reply to the launch of Clawboo. It names the problem all these tools target: Claude Code in one terminal, Codex in another, context copied by hand, and neither agent aware of what the other decided minutes earlier. The comparisons are the author's own, made while promoting his tool.

How it works

Campfire runs Claude Code, Codex, Goose, Aider and OpenHands in one browser tab. Each agent works in its own git worktree, so agents do not edit the same files. Approvals go through permission voting, majority rules, with a 30 second timer, so a single careless approval does not pass unchecked. A shared semantic memory layer carries a fact learned by one agent to the others. Session replay and agent races, where several backends attempt the same task, are also included.

Clawboo takes a different design. A lead agent called Boo Zero splits a request into jobs and delegates them across a mixed team (Claude Code, Codex, Hermes, OpenClaw), which shares one task board and one memory layer. Campfire runs agents side by side under voting. Clawboo uses one agent to direct the others.

Alternatives, as the author describes them

• Conductor: 50 dollars a month, macOS only, capped at three agents.
• Vibe Kanban: free and, in the author's words, "genuinely good". The company behind it shut down, and the post says it has had no commit since April. The year is not stated.
• Crystl: its orchestration commands sit behind a 170 dollar a year tier.
• Clawboo: self-hosted. The post gives no price or licence.

How to use it

bunx the-campfire

This assumes Bun is installed. The repository link in the post is truncated (github.com/stretchcloud/c…), so the full repo name is not visible.

Limits

The post is a launch pitch. It gives no licence name, system requirements, benchmarks or known issues for Campfire. It does not say how votes are cast or what happens when the 30 second timer runs out without a majority. It does not say how the semantic memory is stored or queried. The pricing and maintenance claims about competitors are unsourced and dated to the post, so check them against each vendor before choosing.