Index / MicroClaw · updated Sep 3, 2026

MicroClaw

microclaw/microclaw · healthy · rank 27 of 43 by stars

A self-hosted Rust agent runtime inspired by nanoclaw, unifying one agent loop across Telegram, Discord, Slack, Feishu, and web. Fresh off a v0.5.0 monolith-decomposition sweep with a React 19/Vite 8 operator console.

#rust#agent-runtime#self-hosted#multi-channel#openclaw-alternative
Compare vs OpenClaw
GitHub ↗

Facts

Repository

Contributors27
Open issues1
Last commitSep 3, 2026
Release cadence~4 days
Latest releasev0.5.5

Runtime

LanguageRust
Memory1.8 MB
Boot time8 ms
Deploymentself-hosted · desktop · cloud
Setup difficultyLow
Plugin ecosystemEmerging

Posture

LicenseMIT
Local-firstYes
Cloud dependencyOptional
Multi-userYes
Privacy postureStrong

Community

Sentiment55% positive
Reddit mentions0
Web results9

Security breakdown

Composite 72 / 100 · how these are scored

Sandboxing6 / 10

higher is safer

API security7 / 10

higher is safer

Network isolation6 / 10

higher is safer

Telemetry safety8 / 10

higher is safer

Shell access risk6 / 10

higher is riskier

Model access

Read from the repository, not written by a model · 37 files examined

2 providers · custom endpoint

Direct
AnthropicOpenAI
Compatible
OpenAI-compatible

Pinned models

gpt-5.2OpenAI

claude-sonnet-4-5-20250929 released 2025-09-29 — 11 months old (dated in the model id)

Pin last edited 21 days ago

Evidence

Decision

Why choose MicroClaw over OpenClaw?

Why choose this

  • Rust single-binary runtime with very low resource footprint
  • Channel-independent agent loop avoids per-adapter duplication
  • Built-in durable delivery, resumable sessions, and scheduled tasks

Tradeoffs

  • Much smaller community and skill ecosystem
  • Pre-1.0 with breaking changes on main branch
  • No Reddit presence and thin third-party coverage

Best fit

  • Self-hosters wanting one agent runtime across many chat surfaces
  • Rust shops needing durable sessions and scheduled tasks
  • Operators who want a local control plane with governed extensions

Avoid if

  • You need a battle-tested stable API (main moves fast, pre-1.0)
  • You want a large plugin ecosystem and community
  • You prefer managed cloud hosting over self-hosting
Good Confidence72%

Strong evidence from README, releases, and detailed commit history confirms active development and architecture. Community sentiment is uncertain due to zero Reddit matches and only indirect web mentions.

AI layer reviewed Aug 24, 2026 · how this is written

Star activity

732 stars today

Overview

MicroClaw is a self-hosted agentic AI runtime written in Rust, inspired by nanoclaw's design. Its core architectural bet is a single channel-independent agent loop paired with a provider-independent LLM layer, so Telegram, Discord, Slack, Feishu/Lark, web, and other adapters all share the same behavior without duplicated boundary logic. It targets long-running work: multi-step tool use, resumable sessions, durable delivery, persistent memory, scheduled tasks, MCP, subagents, and governed skill extensions all live in one runtime, fronted by a local control plane at 127.0.0.1:10961.

The recent v0.5.0 release shows serious engineering momentum: a monolith decomposition sweep split multi-thousand-line files (config.rs, db/learning.rs) into domain modules, the web stack moved to React 19/Vite 8, and the operator console got a visual refresh. Commits show disciplined practices—clippy -D warnings gates, blame-ignore revs for large moves, and a stable branch recommended for production while main moves fast.

Compared to OpenClaw, MicroClaw trades ecosystem breadth for a leaner, single-binary Rust footprint and a more opinionated unified runtime. It has a Discord, its own subreddit, and broad i18n docs, but with 732 stars and no independent Reddit discussion, its community and plugin ecosystem remain early-stage.

Nominate a clone

Add a new Claw

Paste a GitHub repository and tell us why it belongs on the tracker.

Opens a prefilled issue on GitHub — every nomination is public. Comfortable with a PR? Adding the repo to projects.json is faster.