Autonomous Agents × Engineering

Deploy autonomous agents in sandboxed Docker containers, coordinated via MCP. Build workspaces with distinct teams and hierarchies, and share them across your devices. Manage agents, delegate tasks, and schedule jobs from anywhere — all encrypted end-to-end with the Signal Protocol.

Agent Workspaces

E2E encrypted

Multi-device

Open Source

Join Waitlist View on GitHub

Built with Rust · Flutter · Python · Docker · MCP · the Signal Protocol

terminal

$ dspatch-cli launch my-workspace

● Parsing dspatch.workspace.yml...

● Building workspace container... [3.2s]

● Starting Docker sandbox... [OK]

● Launching agent: lead (claude-code) [OK]

● Launching agent: coder (claude-code) [OK]

● Launching agent: tester (claude-code) [OK]

Workspace "my-workspace" is running

3 agents active · press Ctrl+C to stop

Scroll to explore

Features

What d:spatch provides

Everything you need to run autonomous agent teams in production — without stitching tools together. d:spatch combines workspace management, agent orchestration, sandbox isolation, and encrypted multi-device sync in a single open-source platform.

Some features listed here are still rolling out — they're coming very soon.

Agent Workspaces

Define your entire agent ecosystem — roles, teams, hierarchies, tools, and runtime parameters in one place. Share, manage and inspect workspaces across devices.

Logging & Observability

Real-time logs, file browsing, and per-agent token usage in one dashboard.

MCP Coordination

Automatic agent discovery, routing, and session management over MCP.

Hierarchies & Teams

Structure agents in multi-level supervisor/worker trees. Supervisors orchestrate sub-agents, delegate subtasks, and manage escalations — while workers focus on execution. Model any engineering workflow as a self-coordinating agent team.

Docker Sandbox

Isolated containers with DinD and full root access. Agents install, build, and run freely — your host stays sealed.

E2E Encryption

All cross-device traffic is end-to-end encrypted. Your data never touches a server you don't control.

File Browser

Inspect what your agents are building. View the files and contents of any workspace from any device.

Scheduling & Delegation

Schedule recurring jobs and delegate tasks to specific agents or teams — all tracked on a shared timeline.

Inter-Agent Communication Protocol

Agents talk to each other directly — one-to-one, one-to-many, many-to-many. Our novel Inter-Agent Communication (IAC) protocol supports arbitrary chat constellations, agent group chats, and even agents interrupting each other.

Inquiry System

Agents can escalate decisions up the hierarchy when decisions exceed their authority. Supervisors auto-resolve these or escalate to higher authorities. Anything unresolved bubbles up as a push notification to your phone, desktop, or any connected device.

Community Hub

Browse, rate, and install individual agents and entire workspaces shared by the community.

Open Source & SaaS

Fully open-source. No account needed. Optional hosted tier for multi-device support.

Mobile Control

iOS and Android via E2E encrypted relay. Monitor and control from anywhere.

Desktop Support

Native app for Windows, macOS, and Linux. No cloud dependency. Deploy, monitor, and control agent workspaces.

Agent & LLM Exposure

Expose any agent locally or across your devices. Works for LLM endpoints as well! The relays are end-to-end encrypted between devices. No port forwarding, no cloud . Just your agents, your hardware — accessible from anywhere.

Agent SDK

Integrate any agent into d:spatch with our Python, Rust, or TypeScript SDKs. Inter-agent communication, exposure, logging, checkpointing and crash recovery are all handled for you.

CLI

Full app control from the terminal. Run headless on servers, script workflows, or manage agents over SSH.

Workspaces

A living agent environment

A workspace is a persistent Docker environment where your agents live and collaborate — running as independent processes, coordinated through MCP in real time.

Everything is defined in a single dspatch.json: agent hierarchy, tool access, environment variables, and schedules. API keys are stored in the app's encrypted local store and referenced by name — nothing sensitive ever touches disk.

my-workspace

■ Stop↺ Restart>_ Terminal

Runningcontainer: a3f2b1cUp 2h 14mTokens: 124kRAM: 2.3 / 8 GB

1 pending inquiry· Coder 1 is awaiting a decisionView

Lead Agentorchestratorrunning

├Coder 1claude-code❓busy

├Coder 2claude-codeidle

├Coder 3claude-coderunning

└Reviewerclaude-codepeers: coderrunning

└Test Runnerpytestbusy

✓Version-controllable & shareable

✓API keys encrypted locally, never on disk

✓Agents run as isolated, independent processes

Communication

Stay in control

Chat with individual agents, answer inquiries as they bubble up the hierarchy, and inspect your workspace files — all from any device, in real time.

Direct messaging

Chat · Coder 1busy

YouWhat's the current blocker?

Coder 1Resolving a merge conflict in auth.ts

Inquiry escalation

Inquiry · bubbling up

You

🔔 push notification on any device

escalated

↻

Lead Agent

no policy — bubbling up

escalated

Coder 1

"JWT or sessions for auth?"

Workspace inspection

Files · my-workspace

📁src/

auth.tsmodified

main.ts

📁tests/

auth.test.tspassing

Security

End-to-end encryption via the Signal Protocol

All data is encrypted on-device using the Signal Protocol (Double Ratchet + X3DH) before transmission. The relay server only ever sees ciphertext.

Your device

Plaintext in

End-to-end encrypted

Relay server

Ciphertext only

Encrypted relay

Your other device

Plaintext out

Per-message keys

Every message is encrypted with a unique key. Compromising one reveals nothing else.

Forward secrecy

Encryption keys rotate automatically. Past messages stay secure even if a future key is compromised.

Trusted device network

Only devices you explicitly verify through a QR code or safety number comparison can join your network.

Agent SDK

Bring any agent into d:spatch

Implement your agent's run loop and the SDK handles the rest — MCP registration, heartbeats, crash recovery, and structured logging come out of the box. Pre-built adapters for Claude Code and other tools mean most agents integrate in minutes, with no boilerplate.

Inquiry escalation via MCP

Agent-to-agent messaging

Usage & token reporting

Structured logging

Persistent agent memory

agent.py

from dspatch import SDK
from claude_agent_sdk import ClaudeAgentOptions, ClaudeSDKClient, ResultMessage

sdk = SDK()

@sdk.agent
async def coding_agent(prompt: str, ctx: sdk.Context) -> str:
    options = ClaudeAgentOptions(
        system_prompt="You are an autonomous coding agent.",
        mcp_servers={"dspatch": ctx.mcp_server},
        allowed_tools=["Read", "Write", "Edit", "Bash", "mcp__dspatch__send_inquiry"],
        permission_mode="bypassPermissions",
        cwd=ctx.workspace_dir,
    )

    async with ClaudeSDKClient(options) as client:
        await client.query(prompt)
        async for msg in client.receive_response():
            if isinstance(msg, ResultMessage):
                return msg.result

sdk.run()

Platforms

Run locally, control from anywhere

One Flutter codebase, native builds for all five platforms. Desktops run workspace containers locally; mobile and secondary devices connect through the encrypted relay for remote monitoring and control.

Desktop

WindowsmacOSLinux

Desktop builds run workspace containers locally via Docker, with full access to the sandbox environment, logs, file browser, and usage dashboard.

Desktop & Mobile

WindowsmacOSLinuxiOSAndroid

Mobile and secondary desktop builds connect to your primary machines through the encrypted relay, providing remote access to workspace monitoring, inquiry responses, and agent management.

Community Hub

Start with what others have built

Browse agent providers, templates, and workspace blueprints shared by the community. Install with a single command — then customize for your project.

Agent providers

Adapters that connect LLMs like Claude, GPT, and Gemini to d:spatch. Install one and your model becomes a managed agent.

Agent templates

Pre-built agents for code review, documentation, testing, monitoring, and more. Deploy in one click, customize to fit your stack.

Workspace blueprints

Full multi-agent setups with roles, hierarchy, and tool access preconfigured. Drop in a team, not just an agent.

Explore

Browse all

The Community Hub is where agents, workspaces, and adapters live.

Explore the hub

Early access

Join the waitlist

d:spatch is in closed development. Drop your email and you'll be among the first to know when the doors open — or dive straight into the source on GitHub.

Get notified at launch

View on GitHub