Distributed, Multiplayer Control Plane for Native Agent Work

Run Codex, Claude Code, DeepSeek Harness, OpenCode, and other native harnesses as isolated workers on an always-on machine. Coordinate people and Agents through durable Projects, Conversations, Tasks, memory, and multi-agent workflows from any connected client.

Website • Hub • Develop • Discord

---

Download the macOS, Windows, or Linux Desktop installer from Releases, launch it, and add the repository you want an Agent to work in. Desktop creates and starts your default local XpressClaw instance automatically—there is no init step.

The matching runner image is pulled automatically.

Why xpressclaw?

Codex, Claude Code, DeepSeek Harness, and OpenCode already supply excellent agent loops. xpressclaw is the control plane around them: durable work, automation, isolation, devices, and a UI for outcomes rather than terminals.

• Multiplayer Project spaces — Coordinate people and specialized Agents through shared Conversations, Tasks, workflows, files, and memory.
• Distributed by design — Keep Agents working on an always-on desktop or server while browsers and future native clients reconnect from other devices.
• Harness-owned intelligence — The selected product owns reasoning, tools, and subagents; xpressclaw is its ACP client, not another agent framework in front of it.
• Structured interface — See tasks, attempts, artifacts, questions, and review decisions without watching a terminal.
• Native desktop app — Tauri installers for macOS, Windows, and Linux with a system tray. Runs in the background, always available.
• Automation-first — Queue tasks, run recurring schedules, and express implementation/review loops as workflows.
• Isolated and continuous — Every Agent reuses one initialized Docker/Podman environment while Project Conversations stay available alongside task work.
• Container-aware workspaces — Opt trusted agents into a separate runner variant with Docker CLI, Compose, Buildx, and access to the host Docker/Podman engine.

Features

Real-time Control Center

Leave the Control center open to see working Agents, active Task and Conversation turns, items needing input, canonical tool calls, context usage, and attributed Git line changes across every Project. One durable live feed replays missed responses and status changes after reconnecting, with direct links back to the relevant work. See the Control center guide.

Project Conversations

Projects are the top-level collaboration boundary. Each Project can contain several specialized Agents, and each Agent can participate in several shared Conversations. Mention one Agent or ask all participants; Conversation turns use independent ACP sessions, so an Agent can answer while its serialized task lane is busy.

Turn a Conversation message into a linked Task with Continue with task, or start a reusable workflow with typed inputs. Agents can create linked Tasks for themselves, publish task results back to the Conversation, and share files that people or other Agents can download into their workspaces.

Codex can also publish interactive Visualize results directly inside Task and Conversation replies. XpressClaw preserves the generated fragment with the message, runs it in an isolated viewer, and asks before an interactive visual can send a follow-up. See Inline visualizations.

Persistent Agents

Each Agent is a durable execution identity inside a Project. It has a name, workspace, retained container, and replaceable ACP harness. Task timelines record every attempt and where it came from, while Project memory and Conversations provide shared context. XpressClaw does not invent a persona or system prompt.

Autonomous Task Execution

ACP harnesses pick up tasks from a queue and publish standard progress, plans, tool activity, and results. Schedule recurring work with cron expressions. Workflows coordinate different Agents—for example, one using Codex to implement and another using Claude to review in a loop—and can report every step back to the Conversation that started the run.

Multiple ACP Harnesses

• Codex: reuses an eligible host ChatGPT/Codex login
• Claude Code: reuses an eligible host Claude subscription login
• DeepSeek Harness: runs through the maintained openma-ai ACP adapter and reuses ~/.dsh
• OpenCode: uses its built-in ACP server
• Custom: any image and command that speaks ACP over stdin/stdout

Each Agent keeps an isolated, reusable environment for its selected harness, workspace, installed tools, and caches. Trusted workspaces can optionally use the host container engine for existing Compose-based workflows.

For DeepSeek Harness, sign in on the control-plane host with dsh-acp login or use dsh web and save a model credential in its Settings page. XpressClaw mounts the resulting ~/.dsh directory only when host login is enabled; no API key belongs in xpressclaw.yaml. See the DeepSeek Harness guide.

Privacy & Safety

• Project boundary, Agent isolation — shared memory and Conversations stay inside one Project; every Agent keeps its own retained container and task ACP process
• Explicit resources — workers receive the configured workspace and volume mounts
• Visible provenance — every queued request records whether it came from a person, schedule, task, or workflow
• Local control data — timelines and task state remain local; agent traffic follows the selected provider's terms
• Credential boundary — subscription auth is mounted only into the configured Agent container built from an image you trust

Full Observability

Session events, attempt lifecycle, artifacts, provenance, and cancellation. Know what ran at 3am and why.

The User Model

• An instance is one long-running XpressClaw control plane plus its local configuration and data. It can own many Projects and repositories. Desktop manages one default local instance at ~/.xpressclaw.
• A client is the Desktop window or a browser connected to an instance. Closing a client does not stop the instance, its queues, or its Agents.
• A Project is the collaboration and memory boundary containing Agents, Conversations, Tasks, and workflows.
• An Agent is an execution identity with one ACP harness, retained environment, and workspace. Its workspace may be an existing repository or an XpressClaw-managed folder.

xpressclaw.yaml configures an instance; it does not add a repository and does not belong in every repository. Repository-local harness instructions such as AGENTS.md, CLAUDE.md, or product-specific equivalents stay with the repository and are read by the selected harness.

Quick Start

1. Install XpressClaw

Desktop (recommended)

The official Releases page publishes these Desktop installers:

• macOS: signed and notarized .dmg files for Apple Silicon and Intel;
• Windows: .exe and .msi installers for x64; and
• Linux: .deb and .rpm packages for x64.

Install and launch the app. It starts the bundled control plane from ~/.xpressclaw, opens first-run setup, and stays available in the system tray. The Desktop installer does not install a separate xpressclaw command on PATH.

CLI/server (macOS and Linux)

Beginning with the first stable 0.3.0 release, install the standalone CLI/server with:

curl -fsSL https://raw.githubusercontent.com/XpressAI/xpressclaw/main/install.sh | sh

The script discovers GitHub's latest stable release at runtime, deliberately ignores prereleases, verifies the published SHA-256 checksum, and installs the single xpressclaw binary to ~/.local/bin. It supports Apple Silicon and Intel macOS plus x64 Linux. Add that directory to PATH if needed, then run:

xpressclaw up

This discovers or creates the default instance at ~/.xpressclaw and opens setup at http://localhost:8935. Windows users should install Desktop; stable Releases also attach a standalone x64 .zip for manual CLI/server installs.

2. Create your first Project and Agent

Choose an existing repository folder or start with an empty managed workspace, name the Agent, and select Codex, Claude Code, DeepSeek Harness, OpenCode, or a custom ACP harness. XpressClaw creates the first Project around that Agent. Add more Projects and Agents from the UI; you do not run init for each repository.

3. Start work

Open a Project Conversation for shared coordination, use Continue with task for durable work, or send private work directly to one Agent from New Work. Closing the window does not stop the local control plane.

Requirements

• Docker or Podman (required for worker isolation)
• At least one supported harness login on the host
• A built-in ACP runner image, or your own ACP-compatible image

Remote Access

The control plane is the durable machine running Agents; Desktop and the web UI are clients. To leave work running on a desktop or server and reconnect from a laptop or phone, keep XpressClaw bound to its default loopback address and put an authenticated transport in front of it.

For example, from a laptop with SSH access to the control-plane host:

ssh -N -L 8935:127.0.0.1:8935 user@control-plane-host

Then open http://localhost:8935 on that laptop. An authenticated HTTPS reverse proxy is another option. XpressClaw does not yet provide native remote authentication, so it refuses non-loopback binds unless --allow-insecure-remote explicitly acknowledges that another security layer protects the address. Never expose the port directly to a LAN or the internet.

Browser disconnection does not cancel work. Reopening the same instance loads durable state and live streams reconnect; after a control-plane process restart, interrupted work is recovered into the queue. See Remote access for the current security boundary and Desktop limitations.

Common Workflows

• Use a Project Conversation to coordinate people and several Agents, or New Work to send private work directly to one Agent.
• Turn larger requests into durable Tasks with progress, artifacts, provenance, results, and errors that remain visible after reconnecting.
• Use Automations → Schedules for recurring or one-off work and Automations → Workflows for multi-Agent implementation/review or goal loops.
• Let the scoped GitHub integration publish a task's pull request, monitor review feedback, and resume the same Agent context until approval or merge.
• Optionally run local GitBucket and Jenkins collaboration services for selected Agents without changing existing GitHub flows.

See the guides for workflows, scheduling, local collaboration, Project management and deletion, remote access, and the Control center.

CLI Essentials

xpressclaw init [INSTANCE]   Optionally provision a control-plane instance
xpressclaw up [--detach]     Start the default local instance
xpressclaw down              Stop its detached control-plane process
xpressclaw status            Show Agent queue status
xpressclaw sync ...          Explicitly fetch/publish portable Project state

Most CLI/server users only need xpressclaw up, status, and down. init is optional; it takes a positional instance directory, while up and down select an alternate directory with --instance:

xpressclaw init /srv/xpressclaw/staging
xpressclaw up --detach --instance /srv/xpressclaw/staging --port 9001
xpressclaw down --instance /srv/xpressclaw/staging --port 9001

The instance directory is the directory containing xpressclaw.yaml; it is not a Project repository or Agent workspace. See the complete CLI command reference and configuration reference for alternate instances, ports, compatibility flags, and YAML settings.

Git-backed Project Synchronization

Use Settings → Project sync to fetch or publish a Project's portable state through a separate Git repository. The CLI supports the same explicit flow:

cd /path/to/platform
xpressclaw sync init --project platform \
  --remote git@github.com:your-org/xpressclaw-data.git
xpressclaw sync publish
xpressclaw sync fetch

Synchronization never runs in the background, and credentials and local runtime settings are not published. For project-name discovery, manifests, the portable/private data boundary, and conflict handling, see the Git-backed Project synchronization guide.

From Open Source to Enterprise

xpressclaw is the open-source foundation. When your team needs collaboration, visual workflows, compliance, and enterprise support — Xpress AI has you covered.

	xpressclaw (Free)	Xpress AI (Enterprise)
Autonomous AI agents	✅	✅
Chat-first web UI	✅	✅
SOPs & scheduling	✅	✅
Local model support	✅	✅
Budget controls	✅	✅
Team collaboration		✅
Visual workflow builder (Xircuits)		✅
iOS & Android apps		✅
On-premise deployment		✅
Role-based access control		✅
Audit logging & compliance		✅
Dedicated support & SLA		✅

Request an Enterprise Demo

Contributing

We welcome contributions. Source-build instructions, architecture, runner development, tests, formatting, and release mechanics are in the Developer Guide.

Community

• Website: xpressclaw.ai
• Hub: hub.xpressclaw.ai
• Discord: discord.com/invite/vgEg2ZtxCw
• Twitter/X: @xpressclaw
• Enterprise: xpress.ai

License

GPL-3.0

---

Built by Xpress AI — the team behind enterprise agent platforms for regulated industries.