DARS

Distributed Agent Runtime System.

An open-source control plane and distributed runtime for coding agents.
Route work across machines, execute with local agent CLIs, and keep every run observable.

Website · Docs · Discord · X · Self-Hosting · Contributing

English | 简体中文

DARS is a Distributed Agent Runtime System. Its Web control plane and Go server coordinate local Daemon/CLI workers, route runs to the right agent runtime, and stream execution progress back to one observable system.

Each task gets explicit runtime configuration, scoped credentials, an isolated work directory, durable events, and a recorded result. DARS is vendor-neutral, self-hosted infrastructure for human + AI teams. It works with Claude Code, Codex, CodeBuddy, GitHub Copilot CLI, OpenCode, OpenClaw, Hermes, Pi, Cursor Agent, Kimi, Kiro CLI, Antigravity, Qoder CLI, and Trae CLI.

For larger teams, Squads add a stable routing layer: assign work to a group led by an agent, and the leader delegates to the right member.

DARS stands for Distributed Agent Runtime System:

• Distributed — the control plane and execution workers can run on different machines.
• Agent Runtime — provider CLIs execute through explicit runtimes, workspaces, credentials, and lifecycle controls.
• System — routing, execution, events, collaboration, skills, and evidence share one durable model.

DARS manages the full distributed agent lifecycle: from task assignment and runtime routing to execution monitoring and skill reuse.

• Agents as Teammates — create agents with explicit runtime, model, skills, environment, and invocation permissions.
• Squads — group agents under a leader agent and assign work to the squad. The leader delegates through comments and mentions.
• Autonomous Execution — set it and forget it. Full task lifecycle management (enqueue, claim, start, complete/fail) with real-time progress streaming via WebSocket.
• Direct Chat and Runs — talk to an agent directly, or create a durable run with separate comments and task execution history.
• Reusable Skills — every solution becomes a reusable skill for the whole team. Deployments, migrations, code reviews — skills compound your team's capabilities over time.
• Local Runtimes — connect local daemons, auto-detect installed agent CLIs, and monitor runtime readiness.
• Multi-Workspace — isolate agents, squads, skills, chats, and runs by workspace.

brew install kms9/tap/dars

curl -fsSL https://raw.githubusercontent.com/kms9/dars/main/scripts/install.sh | bash

dars setup self-host

irm https://raw.githubusercontent.com/kms9/dars/main/scripts/install.ps1 | iex

dars setup self-host

dars setup self-host # Configure, authenticate, and start the daemon

The daemon runs in the background and auto-detects agent CLIs (claude, codex, codebuddy, copilot, opencode, openclaw, hermes, pi, cursor-agent, kimi, kiro-cli, agy, qodercli, qoderclicn, traecli) on your PATH.

Open your workspace in the DARS web app and navigate to Runtimes. Your machine should appear as an active runtime.

What is a Runtime? A Runtime is a local compute environment connected through the daemon. It reports which agent CLIs are available so DARS knows where to route work.

Go to Agents and click New Agent. Pick the runtime you just connected, choose a provider, and configure the agent.

Create a run from Runs (or via dars issue create) and assign it to an agent or squad. The assignee executes it on your runtime and reports progress through comments and task events.

The dars CLI connects your local machine to DARS — authenticate, manage workspaces, and run the agent daemon.

See the CLI and Daemon Guide for the full command reference.

┌──────────────┐     ┌──────────────┐     ┌──────────────────┐
│   Next.js    │────>│  Go Backend  │────>│   PostgreSQL     │
│   Frontend   │<────│  (Chi + WS)  │<────│   (pgvector)     │
└──────────────┘     └──────┬───────┘     └──────────────────┘
                            │
                     ┌──────┴───────┐
                     │ Agent Daemon │  runs on your machine
                     └──────────────┘  (Claude Code, Codex, CodeBuddy, GitHub Copilot CLI,
                                        OpenCode, OpenClaw, Hermes, Pi, Cursor Agent,
                                        Kimi, Kiro CLI, Antigravity, Qoder CLI, Trae CLI)

For contributors working on the DARS codebase, see the Contributing Guide.

Prerequisites: Node.js v20+, pnpm v10.28+, Go v1.26+, Docker

make dev

make dev auto-detects your environment (main checkout or worktree), creates the env file, installs dependencies, sets up the database, runs migrations, and starts all services.

See CONTRIBUTING.md for the full development workflow, worktree support, testing, and troubleshooting.

An iOS mobile client lives in apps/mobile/ — see its README for how to build it onto your own iPhone.

Multica License — the complete Apache License 2.0 text incorporated together with additional conditions — see NOTICE for attribution notices.

• Providing Multica as a hosted service to third parties, or embedding it in a commercially distributed product, requires a commercial license obtained from the producer (condition 1a).
• Unless the producer has granted a written branding waiver, the Multica LOGO, product name, and copyright information may not be removed or modified in a Multica user interface. The user interface is defined by derivation — including apps/web/, apps/desktop/, apps/mobile/, packages/views/, and packages/ui/ — and covers raw source, the frontend container image, and compiled desktop and mobile binaries (condition 1b).
• Non-interface use (running only the server/ backend, the daemon, or the CLI) is exempt from the branding condition, but must retain the source and NOTICE attribution and state that the product is built on Multica, with a link back to this repository (condition 1c).
• A branding waiver and a commercial license are separate grants; neither implies the other (condition 1d).