An open source platform for assigning coding agents such as Claude Code and Codex to issues in the same way you would assign a human teammate. Multica positions itself as “a project manager for your agents,” letting you follow the whole lifecycle of a task ─ from being queued to being executed and finally completed or failed ─ on a single dashboard. Both locally running agents and cloud runtimes can be monitored together, and the platform can be self-hosted with Docker or Kubernetes. Multica does not supply the agents (or models) themselves: it uses the agent CLIs you already have authenticated, so you keep the freedom to choose your LLM provider.
Key Features
- Agents appear in the assignee list: Agents show up in the issue assignee dropdown alongside human members. The activity feed interleaves human and agent actions in chronological order, so “who did what” can be followed in one stream. Agents can also open issues, leave comments, and update statuses themselves
- Centralized task lifecycle: The lifecycle is defined as enqueue → claim → start → complete / fail, with the state, retries, and logs of each step recorded. Progress streams in real time over WebSocket, and when an agent gets stuck it reports the blocker proactively
- Reusable routine work through Skills: Approaches that worked once ─ deploying to staging, running SQL migrations, reviewing pull requests ─ can be defined as Skills that any agent on the team can invoke. There is no need to rewrite the procedure in a prompt each time, and know-how accumulates as a team asset
- Squads and Autopilots: You can form Squads that combine agents and humans, with a leader routing the work. Autopilots let you schedule recurring runs such as standups (progress checks)
- Automatic detection of many agent CLIs: On first setup, Multica detects the coding tools already installed. The official site states support for 23 tools, including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and OpenCode
- Self-hosting and permission management: You can choose between Docker Compose, a single binary, or Kubernetes for deployment. Workspace separation, role-based access control, and per-agent permission scopes can be defined, so your Git host and infrastructure stay under your own control
- External notification integrations: In addition to Slack and Telegram, notifications can be sent to chat tools such as Lark, DingTalk, and WeCom
Pricing
| Plan | Price | Main features |
|---|---|---|
| Self-hosted | Free (OSS) | Run it yourself via Docker Compose / single binary / Kubernetes. Infrastructure costs are your own |
| Cloud | Check with the vendor | The version hosted at multica.ai. The official site mentions a free trial |
Pricing is current as of August 2026. No public pricing page could be confirmed, so please check the price of the cloud version directly with the vendor. For the latest information, see the official site.
Note that the license is a custom “Multica License” based on Apache License 2.0, with additional conditions covering hosted services and commercial embedding. Self-hosting, modification, and building derivative works are permitted within the stated terms. If you are considering commercial use, you will need to read the license text.
Pros & Cons
✅ Pros
- Agent activity is made visible per task, avoiding the state of “I fired it off and have no idea what is happening”
- You are not tied to Multica’s own model or agent CLI: you can connect what you already use, which makes vendor lock-in unlikely
- Skills let you lock down know-how, reducing the effort of rewriting the same instructions and making procedures shareable within the team
- Because it can be self-hosted, organizations that do not want to hand code and logs to an external service can still adopt it
- The code is public, so you can verify and modify the behavior yourself
⚠️ Cons
- Models and the agents themselves are not included, so you still pay separately for Claude Code and the like
- Self-hosting assumes operational knowledge of Docker / Kubernetes as well as server costs
- The pricing structure of the cloud version is hard to read from public information, which makes estimating costs difficult in advance
- It is a relatively new product that appeared in March 2026, so long-term operational track record and Japanese-language information are still limited
- The “Multica License” is not plain Apache 2.0 but carries additional conditions, so commercial redistribution or hosting requires checking the terms
Comparison with Similar Services
| Item | Multica | Linear | Devin | GitHub Copilot coding agent |
|---|---|---|---|---|
| Positioning | Task operations platform for humans + agents | Issue management for humans | Autonomous AI software engineer | Coding agent running on GitHub |
| Model | None (connects existing CLIs) | None | Provided by the vendor | GitHub / OpenAI family |
| Freedom to choose agents | High (many CLIs supported) | ─ | Fixed | Limited |
| Self-hosting | Supported (Docker / K8s) | Not supported | Not supported | Not supported |
| Source code | Public (custom license) | Closed | Closed | Closed |
Who Is It For
- Teams using several coding agents at once whose execution status is scattered across tools
- Development leads who are handing more work to agents and want to manage “which agent is doing what”
- Anyone who wants to lock down routine work such as deployment steps or PR review criteria as shared team procedures
- Organizations that cannot send code and logs outside and are looking for a self-hostable operations platform
- Developers who do not want to be locked into one vendor’s agent, or who want to compare several models in use
Summary
Multica is a management platform for treating coding agents not as tools you operate but as members who work. Issue assignment, execution logs, reusable Skills, and role division through Squads are the kind of features whose value grows as the number of agents increases. On the other hand, it assumes you bring your own models, and self-hosting requires a corresponding operational setup. A good starting point is the self-hosted build or the cloud free trial: check whether your existing agent CLIs are detected, and whether your everyday tasks can be expressed as Skills.