open-mercato/cezar

Open-source orchestrator and ADE for running Claude Code, Codex, OpenCode, Pi and other AI coding agents in parallel — locally or 24/7 on your own server.

What it solves

Cezar is a control center for managing multiple AI coding agents (such as Claude Code, Codex, and OpenCode) in parallel. It solves the problem of running agents locally on a laptop where work stops when the laptop closes, and prevents agents from interfering with each other by giving each task its own isolated git worktree.

How it works

Cezar acts as an orchestrator that wraps around existing agent CLIs. Users describe a task or start from a GitHub issue, and Cezar runs a predefined workflow (defined in YAML) consisting of agent steps and shell checks. It streams the agent's progress live to a web-based cockpit. If a shell check fails, the agent can automatically retry. Agents can also delegate independent work to child tasks, each in its own worktree, and report results back to the parent.

Who it’s for

Developers who use AI coding agents and want to automate multi-step workflows, run tasks 24/7 on a VPS, and manage multiple projects from a single dashboard.

Highlights

  • Parallel Execution: Each task runs in its own git worktree, allowing multiple agents to work simultaneously.
  • Autonomous Mode: Allows agents to complete tasks without stopping to ask for user permission.
  • Workflow Automation: Uses YAML files to chain agent steps and shell commands for verification.
  • Agent Agnostic: Works with various agent CLIs (Claude Code, Codex, Copilot, etc.) using existing logins rather than API keys.
  • External Integrations: Connects to GitHub, Jira, and Linear to launch tasks directly from issue trackers.
  • Cezar Cockpit: A responsive web interface for monitoring live runs, comparing diffs between different agent variants, and tracking token costs.

Related

  • Project
    liustack/modsearch

    ModSearch is a free‑first plug‑in that gives LLM chat apps (DeepSeek, Claude Code, Codex, etc.) live web, page‑fetch, and X (Twitter) search capabilities. It routes requests through engines like Firecrawl (key‑less default), Antigravity CLI, Tavily, Exa, Grok Build, or a local fetcher, with automatic fail‑over and key rotation. Installable via a single command, it works out of the box and returns structured, sourced answers.

  • Project
    Waikato/moa

    MOA is an open-source Java framework for real-time data stream mining and large-scale machine learning, providing a suite of algorithms and evaluation tools.

  • Project
    QuantClimate/GPJax

    A low-level Gaussian process framework built in Jax that provides researchers with a flexible, mathematically-aligned interface for building and extending GP models.

  • Project
    ganeshmshetty/openclip

    An open-source macOS text utility that provides a contextual popup bar with AI tools, text transformations, and custom extensions for selected text.