Multica, an open-source managed agent platform, launches, turning coding agents into team members

Multica puts coding agents and human members in the same workspace: after you name an agent and specify its runtime and model provider, it appears on the task board like a colleague. Users assign issues to the agent, which picks them up on its own, executes on a machine you control (a daemon on your local or cloud host), comments along the way, and moves the issue to review when done. The platform ships no models; it drives the agent CLIs you've already installed and logged into, with an official list of 26 supported—including Claude Code, Codex, Cursor, GitHub Copilot CLI, Grok, Kimi, and Qwen Code—and switching providers is just a dropdown change. For observability, every run's tool calls, commands, and errors are timestamped and replayable, with per-run cost viewable by agent and issue; you can interject mid-run and messages land in the current run rather than the next (currently supporting Claude Code, Codex, and Grok). Failed runs auto-retry or stop with an explanation, and output goes to review rather than straight to main. There are also Autopilots scheduled tasks, Skills reuse, Squads grouping, and integrations with Slack, Feishu, DingTalk, WeCom, Telegram, and more. Deployment supports cloud, desktop, and self-hosting: self-hosting uses Docker Compose or Helm, requires Docker, and connects to GitHub, GitLab, Gitea, and Forgejo (including self-hosted); the stack is Next.js 16, a Go backend, and PostgreSQL 17. The security model defines what agents can and cannot access, with an audit trail. Self-hosted servers send one anonymous deployment-level snapshot per day (version plus grouped counts, no names, content, or IDs), which DO_NOT_TRACK=1 disables. Prerequisite: at least one supported agent CLI installed and logged in on the machine running agents. Licensed under the Multica License—the full Apache 2.0 text plus additional conditions for hosted services, commercial embedding, and branding.