multica-ai/multica

Multica is worth a look if you already run several coding-agent CLIs and spend too much of the day switching terminals and repeating context. It is a coordination layer, not a model or an autonomous coder.

The problem

You have Claude Code in one tab, Codex in another, and each session forgets what it knew. Cross-issue context, progress notes and the eventual diff live in separate places, so you keep re-explaining the same setup. Multica changes the entry point: you file an issue and assign it to an agent, and that agent picks it up on a connected runtime, comments as it works, and hands the result back for review. The intent, the run log and the diff stay attached to one issue.

What's different

The core mechanism is the runtime: Multica does not ship a model, it drives agent CLIs you install and authenticate yourself, and the README currently lists 26 supported providers including Claude Code, Codex, Cursor Agent, Copilot CLI, Kimi and OpenCode (runtimes). Switching provider is a configuration change rather than a migration. Each agent gets a name and runtime and appears on the board like a teammate; squads let a leader route work across people and agents (squads).

The practical payoff is traceability. Every run has a timestamped execution log of tool calls, commands and errors (tasks), usage analytics split cost per agent and issue, and work lands in review instead of main. You can also reply to a running agent and have the message land in the current run, though the README limits that to Claude Code, Codex, Grok and Kiro. The trade-off is that Multica adds a workspace and a runtime daemon on top of tools you already trust; it does not reduce the supervision needed, it reorganises it. A commenter on the Hacker News thread made the same point about the "teammate" framing: an agent may fail confidently and silently, so someone still has to watch the output (comment, 2026-04-11, opinion from that thread).

Who should try it

Good fit: a small team already paying for two or more agent CLIs and working through an issue tracker, who want assigned work, review gates and per-run cost in one place. A single developer with one agent and one repository will likely find the extra moving parts outweigh the coordination benefit. Multica also offers a hosted cloud signup and a desktop app, so you can evaluate the workflow without standing up infrastructure yourself (desktop download, cloud quickstart). The self-hosted path matters if your code and Git host cannot leave your network.

Requirements and limitations

The one hard prerequisite: the machine that runs agents must have at least one supported agent CLI installed and signed in, because Multica drives those tools rather than shipping them. On a self-hosted deployment you also run the full stack yourself via Docker Compose or Helm (self-hosting), which is ongoing work, not just an install. Git integration covers GitHub, GitLab, Gitea and Forgejo, including self-hosted instances (VCS integration), so check yours is on that list before committing. The license is source-available: Apache 2.0 plus additional conditions on hosted services, commercial embedding and branding (license). That is a real constraint for anyone planning to resell or embed the workspace.

HotRepo's take

HotRepo tracked 6 daily-board days between 2026-05-21 and 2026-10-27, from about 30.5k stars to 54.0k; the most recent recorded day added +123 (project). Those snapshots show visibility, not production adoption, and the gaps prevent treating them as a growth curve. Before investing, verify the specific CLI and Git host you depend on are in the documented support lists, and read the license conditions for your intended use.