Multica

Assign issues to agents and keep execution, decisions, and review attached to the same work item

Multica lets agents be assigned to issues much like human teammates. An assigned agent picks up the task on its runtime, posts progress or blockers, and returns completed work for review. The design keeps intent, runs, decisions, and diffs connected to the issue instead of scattering context across separate terminal sessions.

Sources: [1]

Connect 26 agent CLIs through runtime daemons that execute beside the code on machines you control

The official README lists 26 CLI runtimes including Claude Code, OpenAI Codex, Cursor Agent, GitHub Copilot CLI, OpenCode, Kimi, and Grok. Multica does not bundle a model or coding agent. Its daemon discovers agent CLIs that are already installed and authenticated on a machine, then executes assigned tasks on that runtime.

Sources: [1][2]

Track execution logs, token usage, retries, timeouts, and human review gates for agent work

Runs expose timestamped tool calls, commands, and errors, while token usage can be reviewed by agent and issue. Automatic retries and timeouts, an Inbox for decisions that need human input, and review gates after completion create explicit observability and approval points around autonomous work.

Sources: [1]

Expand from individual tasks into squads, reusable skills, scheduled autopilots, Git hosts, and chat channels

Humans and agents can be grouped into squads, solved procedures can become reusable skills, and standups, audits, or reports can run as scheduled autopilots. Repositories can come from GitHub, GitLab, Gitea, or Forgejo, while team channels such as Slack or Lark can trigger and follow agent work. The control plane can be self-hosted with Docker Compose or Helm.

Sources: [1][2]

For teams that already use several coding agents but need shared ownership, review, and visibility

Multica fits teams that use Claude Code, Codex, or other coding agents for different jobs but lack a common place to see which agent owns an issue, which run produced a change, and where a human must review it. It is best understood as an assignment, runtime, audit, and review layer on top of existing agent CLIs rather than another agent model.

Running beside the code does not remove the separate data boundary of each agent CLI and model provider

The Multica daemon executes agent CLIs on the machine that holds the code, but Multica does not provide the model or the agent CLI itself. Whether Claude Code, Codex, or another runtime sends prompts or code context to an external provider depends on that provider and configuration. With Multica Cloud, control-plane data such as issues, runs, and comments also lives in the hosted service, so evaluate the code execution location separately from workspace metadata storage.

A self-hosted deployment moves the Go backend, Next.js frontend, and PostgreSQL database to infrastructure you control. Production authentication still needs appropriate email or identity configuration, and the self-hosting guide warns against exposing a deterministic development verification code publicly. Every runtime machine also needs a supported agent CLI installed and authenticated separately.

Sources: [1][2]

The Multica License is not plain Apache-2.0 and restricts third-party hosting, commercial embedding, and branding

The LICENSE combines the full Apache License 2.0 text with additional Part I conditions. Without a commercial license, it restricts using the Multica source to provide a hosted service to third parties or embedding it in a commercially distributed product or service, while internal use within one organization is allowed. A public instance for users outside the organization requires a commercial license even if it is free.

The license also preserves conditions around the Multica name, logo, and attribution, with documentation attribution requirements for backend, daemon, or CLI uses without the Multica UI. GitHub therefore reports NOASSERTION rather than Apache-2.0. Teams should evaluate the complete Multica License rather than treating the repository as standard Apache-2.0 software.

Sources: [3]