dispatchmy.ai
How it works Compare Docs FAQ Sign in Join waitlist
Comparison

OpenClaw vs dispatchmy.ai

OpenClaw is a popular open-source AI agent that runs as a process on your machine. dispatchmy.ai is a contained team of specialist agents you configure from a dashboard. Here's how they compare.

TL;DR

One sentence each.

OpenClaw
Maximally capable AI assistant that runs on your machine, talks across every messaging platform, and can rewrite its own skills.
Right for: technical tinkerers who want maximum capability, accept broad system access, and are comfortable with an agent that modifies itself.
dispatchmy.ai
Managed team of specialist agents in containers, configured by a dashboard. Reliable across updates, predictable across sessions.
Right for: people who want a reliable, predictable specialist team without becoming a sysadmin or trusting the agent with their machine.
Side by side

The differences that matter.

Three architectural choices drive everything else. Both products ship multi-agent specialists; the table below skips the overlap and focuses on where the platforms genuinely diverge.

ConcernOpenClawdispatchmy.ai
Where the agent runs By default, as a process on your machine — reach bounded only by your user account's filesystem permissions. An opt-in Docker sandbox can confine tool execution to a container, but it's off out of the box and the gateway always stays on host.Sandbox docs hedge: "not a perfect security boundary, but materially limits filesystem and process access." Inside an isolated container by default, per workflow. Reach is bounded by explicit volume mounts.Same capabilities as on host, smaller blast radius. Recovery = restart the container.
Who configures the agent You, by editing markdown files (AGENTS.md, SOUL.md, TOOLS.md) — or the agent itself, by writing to those files.Two failure modes: filesystem gatekeeping for non-technical users, plus agent-induced corruption. You, in a dashboard. The agent has no path to its own configuration.Predictable, reviewable, recoverable. Updates can't be clobbered by the agent.
Self-modification First-class feature. Agent rewrites its own prompts and skills with hot-reloading.Powerful, but documented to break itself; updates can overwrite customizations. Off by default. You configure agents from the dashboard; the agent itself doesn't have a path to its own setup, so updates can't silently clobber it.Trade: less autonomy in exchange for "same input, same agent, every time".
Distribution model Open-source, MIT. You self-host, you self-support, you self-maintain. Closed-source. Local install via Docker, cloud-managed updates.
Specialist agents 9 prebuilt specialists with one-command setup, plus subagent spawning via tool calls. Build your own specialist team — composable into hierarchies of any depth. Prebuilt templates on the roadmap.
Composition model Subagents are ephemeral background runs scoped to a parent session (agent:<id>:subagent:<uuid>). Hard depth cap — default 1, max 5. Leaf subagents lose key tools by default.A "subagent" isn't a reusable entity — it's an inline-spawned run from a parent's allowlist. Every agent is first-class — referenceable as a tool by any other agent, in trees of any depth, with the same tools and identity wherever it sits.No "subagent" type. The same agent you ran standalone yesterday is the same agent another agent calls today.
Channels WhatsApp, Telegram, Slack, Discord, Signal, iMessage, WebChat — all routed through one Gateway. Telegram today, plus the dispatchmy.ai dashboard. More channels on the roadmap.
Prompt-injection blast radius Default install: an injection inherits your user account's reach on the host — read or exfiltrate your home directory, delete files, even wipe the filesystem.With sandbox enabled (opt-in): tool damage is contained to a Docker container, but the gateway still runs on host, tools.elevated is an intentional sandbox-bypass, and the project's own docs decline to call the sandbox airtight. Damage is capped at the container's mounted scope by default. The agent doesn't dynamically grant itself new tools or permissions.
When to pick which

Honest framing — pick the one that fits.

Pick OpenClaw if you…

  • Are security-conscious — willing to enable the sandbox, audit tool grants, and own the threat model yourself
  • Are comfortable editing markdown config files
  • Want messaging-channel-first access (Telegram, Discord, etc.)
  • Prefer open-source you can fork and modify
  • Accept that the agent may modify itself, and that updates may overwrite changes

Pick dispatchmy.ai if you…

  • Want safe defaults — no sandbox to configure, no threat model to own
  • Want a contained, predictable team of specialist agents
  • Prefer to configure by clicking, not by editing files
  • Need behavior that doesn't drift across updates or sessions
  • Want a vendor on the hook for support and updates

Want AI agents you can actually trust on your machine?

If "managed, contained, dashboard-configured" sounds right, we're the team. Early access is in waves — bring your own LLM key.

Join waitlist →