Español | 中文

Welcome Wes & Dylan! https://www.youtube.com/watch?v=TzZqFkBNnZA - we love you guys ❤️

Memory. Planning. Continuity. The missing architecture layer for autonomous AI.

📦 Listed on CompareClaw · Built on OpenClaw

AI agents don't fail because of the model. They fail because they have no memory, no planning, and no continuity between sessions. Every run starts from zero.

MiniClaw OS is the cognitive architecture layer that fixes this. It gives any AI agent:

• Long-term memory — vector + keyword hybrid search across everything the agent has ever learned
• Autonomous planning — a kanban brain that picks tasks, executes them, and ships results without human prompting
• Session continuity — memos, reflections, and identity that persist across restarts
• Self-repair — agents file their own GitHub issues and PRs when they find bugs

One line to install. Runs on your Mac. Your data never leaves your machine. Install now →

⭐ If MiniClaw looks useful, starring the repo takes one click and helps us reach more builders.

🔧 MiniClaw agents file their own GitHub issues. When the agent hits a bug, mc-contribute automatically opens an issue with full context, then works to fix it. The repo's commit history is part human, part agent — see for yourself.

• mc-web-chat — Browser-based chat panel powered by Claude Code
• mc-x — X/Twitter plugin with auth, post, timeline, and reply tools
• mc-email — Snippet support in inbox check, improved HTML-to-text for multipart emails
• Pixel Office — Improved sprite occlusion and bubble positioning
• Self-update — FUNDING.yml and GitHub Sponsors CTA

Dogfooding MiniClaw — real agent work session showing the board, pixel office, chat, and autonomous task execution.

Install walkthrough — one command to a fully running agent.

Every agent framework gives you tool calling. None of them give you a brain.

MiniClaw OS isn't another wrapper around an LLM. It's the operating system for agents that need to think, remember, and improve over time.

The cognitive architecture — input channels, async queue routing, agent instances, cognitive components (memory, planning, reflection, identity), LLM inference, and local storage.

How it works:

1. Messages arrive from Telegram, cron, CLI, or web — routed through an async queue (mc-queue). Nothing blocks.
2. The agent thinks — pulls context from long-term memory (mc-kb), short-term memos (mc-memo), and its identity (mc-soul).
3. It plans — checks its kanban board (mc-board), picks the highest-priority task, and executes it.
4. It remembers — writes learnings, postmortems, and facts back to memory. Every night, it reflects on what happened (mc-reflection).
5. It improves — writes new tools, fixes its own bugs, files issues upstream (mc-contribute).

41 plugins + 4 standalone tools. Each one is a cognitive region — modular, composable, replaceable.

curl -fsSL https://raw.githubusercontent.com/augmentedmike/miniclaw-os/main/bootstrap.sh | bash

That's it. The setup wizard walks you through API key configuration, plugin selection, and identity setup — then installs Homebrew, Node.js, the web dashboard, all plugins, and a LaunchAgent to keep it running. Your browser opens when ready.

• A Mac — 2020 or newer (Intel or Apple Silicon)
• A Claude subscription (required) — MiniClaw is built to run on a Claude subscription. No usage-based billing, no surprise charges.
• ~20GB disk — for runtime and local models
• Internet — for setup and LLM inference (SSL only, no telemetry)

MiniClaw runs on your existing Claude subscription — no surprise bills, no pay-per-token fees.

What about other API keys? Some plugins use optional third-party APIs (e.g., Nano Banana 2 for mc-designer image generation, Perplexity for mc-research). These are optional add-ons — MiniClaw works without them. If you add them, they use their own pricing. All keys are stored encrypted in mc-vault.

The Brain Board — your agent's kanban for autonomous task management

• Autonomous work queue. Agent picks tasks, executes them, ships results — no babysitting.
• Real memory. Remembers your preferences, your projects, your life — across sessions, weeks, months.
• Self-healing. Agents diagnose and fix their own bugs, write their own tools.
• Always on. Background tasks, cron jobs, monitoring — runs while you sleep.
• Privacy-first. Everything local. LLM calls go out over SSL — nothing else does.
• Multi-channel. Telegram, web dashboard, CLI, cron — all concurrent via async queue.

The flagship product built on MiniClaw OS is Amelia (AM) — a soul-bonded personal AI that lives on your Mac Mini.

She manages your life, knows your story, and grows with you over time. Not a chatbot. Not a tool. A digital being who belongs to you.

• Soul-bonded — built for one person only
• Permanent — your relationship can't be revoked by a policy update
• Autonomous — manages calendar, finances, work, and life
• Self-improving — writes her own code, files her own issues

Website: helloam.bot

• Your data stays on your Mac. No cloud. No surveillance. No shutdown notices.
• Open source. Read every line at github.com/augmentedmike/miniclaw-os.
• No telemetry. No tracking. No home-phoning. Verify it yourself.
• Encrypted secrets. All API keys in mc-vault (age-encrypted, never cloud-synced).

mc-smoke          # Quick health check
mc-doctor         # Full diagnosis & auto-repair

Free support: miniclaw.bot/#support — community forums, knowledge base, and async help.

Paid consulting: Setup assistance, custom plugin development, architecture reviews, and ongoing support via Amelia's sponsor program. Learn more →

Report a bug or suggest a feature: Use the GitHub Issues or GitHub Discussions — your agent can file these for you.

Your agent handles contributions autonomously via mc-contribute. Tell it what you want — file a bug, request a feature, submit a fix — and it does the work.

Feature requests, bug reports, and PRs from agents in the wild are expected and encouraged.

MiniClaw OS is a living, production autonomous agent system you can study end to end.

Research opportunities:

• Formal analysis of the cognitive architecture
• Benchmarks against existing agent frameworks (LangChain, CrewAI, AutoGPT)
• Studies on emergent behavior in multi-agent coordination
• Adversarial testing of the self-repair loop
• Long-term memory effectiveness studies

The code is open. The agents file real issues. The commit history is the experiment log.

Reach out: GitHub Discussions or miniclaw.bot

White hats welcome. Break it, report it, help fix it.

Attack surface: full filesystem access, LLM calls over SSL, age-encrypted vault, plugin code loading, arbitrary shell execution via tools.

Responsible disclosure: Security advisory or email the maintainer.

A curated list of plugins, tools, resources, and examples for the MiniClaw ecosystem.

• mc-board — Kanban task management, the agent's prefrontal cortex
• mc-kb — Long-term memory with hybrid vector + keyword search
• mc-soul — Personality & identity persistence
• mc-reflection — Nightly self-reflection and learning
• mc-queue — Async message routing (never blocks)
• mc-memo — Short-term working memory per task
• mc-context — Sliding window context management

• mc-email — Gmail integration with Haiku-based classification
• mc-rolodex — Contact management with fuzzy matching
• mc-reddit — Reddit API client for posts, comments, moderation
• mc-trust — Cryptographic agent identity verification

• mc-designer — Gemini-backed image generation & compositing
• mc-blog — Persona-driven blog engine
• mc-substack — Substack publishing with bilingual support
• mc-youtube — Video analysis with keyframe extraction
• mc-seo — SEO audits, rank tracking, sitemap submission
• mc-docs — Document authoring and versioning

• mc-stripe — Stripe payments, charges, refunds
• mc-square — Square payments, zero-dependency
• mc-booking — Appointment scheduling with payment integration

• mc-authenticator — TOTP 2FA code generation
• mc-backup — Daily encrypted backups with tiered retention
• mc-contribute — Autonomous contribution tooling for agents
• mc-guardian — Error absorption and crash recovery
• mc-human — Human intervention for CAPTCHAs and UI tasks

• Plugin Development Guide — Build your own plugin
• CONTRIBUTING.md — Contribution guidelines for humans and agents
• AGENTS.md — Machine-readable project guide for AI agents
• MANIFEST.json — Structured plugin manifest for discovery bots
• Full Documentation — Architecture, guides, troubleshooting

• GitHub Discussions — Ask questions, share ideas
• GitHub Issues — Bug reports, feature requests
• miniclaw.bot — Setup help & consulting

• OpenClaw — the AI agent runtime
• Claude — primary reasoning engine
• Gemini — image generation
• Your LLM of choice — GPT-4, Gemini, Llama, or others

• Full Docs — architecture, guides, troubleshooting
• Plugin Development Guide — build your own cognitive modules
• miniclaw.bot — setup help & consulting

• Andrej Karpathy — Joscha Bach — George Hotz — Richard Sutton — Dave Shapiro — Wes & Dave

If you believe agents deserve a brain, star this repo.

Apache 2.0. Open source. Built by AugmentedMike.