One embeddable Rust agent core for Server, native Work, and your own apps.
Use MicroClaw as an always-on Server, a native local Work app, or an SDK inside your own Rust product. Every path shares the same tools, memory, Skills, policy, and run lifecycle.
For Rust product and platform teams
Embed the complete Agent Engine—or only its stable lifecycle contracts—inside your own application.
microclaw-sdk = { version = "0.6.1", features = ["full"] }For people working locally with projects
Install the native desktop experience and start a workspace-aware agent conversation in minutes.
brew tap microclaw/tap && brew install --cask microclaw-workFor services, channels, and automation
Run a durable agent service for chat channels, Web and agent APIs, schedules, and background work.
curl -fsSL https://microclaw.org/install.sh | bashOne shared Engine for Server, Work, and embedded apps
AGENTS.md and structured SQLite memory with observability
Tool loop, native and ACP subagents, scheduling, and background work
Skills catalog and MCP tool federation
A runtime-centered stack that keeps your logic stable while channels, models, and tools evolve.
Server, Work, and third-party apps use the same Agent Engine through stable run contracts.
A Rust/GPUI workspace projects shared runtime events without bundling the Web UI.
The model chains tools across multiple steps until the task is complete.
A Main Agent can delegate observable, cancellable work to bounded native or ACP workers.
File and structured SQLite memory include extraction, deduplication, and lifecycle controls.
Built-in skills and external MCP servers extend the runtime without changing the core loop.
Usage, memory observability, config checks, metrics, and traces make drift visible.
Recurring and one-shot tasks run through the same runtime.
Server, Work, and third-party Rust apps converge through the SDK on one provider-neutral Agent Engine and Runtime.
Agent Loop · Run lifecycle · Events · Recovery
Use Server for channels, Work for native desktop tasks, or microclaw-sdk inside a Rust app.
Every entry point calls the same Agent Loop, providers, Skills, tools, policy, and memory.
Server delivers channel events while Work maps the shared stream into durable desktop state.
Sessions, checkpoints, approvals, memories, and task state retain consistent semantics.
Server supports macOS, Linux, and Windows. Work supports macOS; Linux and Windows portable previews are available.
From individual operators to platform teams, one runtime covers multiple paths.
Use native Work conversations, workspace context, checkpoints, and approvals on macOS.
Read implementation detailsRun one Server across chats with memory, scheduling, and shell or file tools.
Read implementation detailsUse permission-aware tools and session history for recurring internal workflows.
Read implementation detailsAdd a skilled Agent and ordered runtime events to a Rust app without importing Server or UI code.
Read implementation detailsStart with setup, then explore tools, permissions, memory, and channel deployment.