Self-Host Multica: Orchestrate AI Coding Agents as Teammates
Multica's open-source platform manages Claude Code, Codex, and similar agents in shared workspaces with full self-hosting via Next.js/Go/PostgreSQL stack and local daemons—no Multica Cloud required.
Architecture Enables Distributed Agent Execution
Multica separates management from execution: a Next.js frontend, Go backend, and PostgreSQL 17 with PG Vector handle workspaces, boards, issues, agent profiles, reusable skills (e.g., deployment flows), real-time updates, and task assignments. A local agent daemon runs on user-controlled machines (laptop, Mac mini, Linux box), auto-detects installed CLIs like Claude Code, Codex, OpenClaw, or OpenCode, and executes tasks assigned via the web app. This keeps coding work on your hardware while centralizing oversight, supporting multiple repos/agents/people without vendor lock-in—agents register runtimes dynamically for heterogeneous setups.
Self-Hosting Avoids Cloud Lock-In with Simple Local Setup
Skip default install (which uses Multica Cloud); use make selfhost after cloning repo or multica setup --local. This generates .env/JWT secret, spins up Docker Compose (frontend: localhost:3000, backend: 8888), and enables local login with any email + code 888888. Install Multica CLI + agent CLI on runtime machines, run multica config local, multica login, multica daemon start (or one-shot multica setup --local). Verify in app settings > runtimes/agents: create agent, assign to workspace/issue—it auto-picks up tasks. No external auth; browser talks to your localhost.
Production requires custom domain/TLS (Caddy/Nginx examples), external PostgreSQL 17/PG Vector, env tweaks (API/WebSocket URLs). Email auth uses Resend API key/magic links (stdout fallback if unset; Google OAuth optional). File storage: local default or S3/CloudFront. Point CLI app/server URLs to your domains (e.g., app.example.com)—login stays self-contained. Upgrades: git pull, rebuild Docker, auto-migrations on backend start.
Structured Workflows Scale Teams, But Weigh Costs
Agents post updates, report blockers, share boards with humans—like Jira for AI teammates. Reusable skills compound (e.g., code review patterns persist). Ideal for multi-repo/agent coordination; overkill for single-repo solo use (stick to direct CLI). Trade-offs: infrastructure costs (DB, domains, TLS), potential Resend/S3 fees, underlying model expenses (Anthropic/OpenAI). Not air-gapped—depends on agent CLIs/models. Vendor-neutral daemon beats cloud silos, but demands setup effort for control/flexibility.