Tinyclaw
The original Tinyclaw as your personal autonomous AI companion. 🐜
Last verified:
What is Tinyclaw?
Tinyclaw is a personal, self-improving, self-learning autonomous AI companion built from scratch as a native framework with zero dependency on external AI frameworks. It is designed to be simple, affordable, and truly personal—like having your own Codsworth or AYLA as a helpful friend rather than a robot slave. The framework features a tiny core with plugin architecture, where channels, providers, and tools are all plugins that you add only as needed.
Key features include a Discord-like dark-themed web UI with real-time SSE streaming and typing indicators, a 3-layer self-improving memory system (episodic, semantic FTS5, temporal decay), smart 8-dimension query routing that cuts LLM costs by tiering queries across providers, a Heartware personality engine with SOUL.md and IDENTITY.md, context compaction with 4-layer pipeline, delegation system for autonomous sub-agent orchestration, 5-layer security with path sandbox and rate limiting, and Bun-native runtime for maximum performance. It supports multi-provider fallback (Ollama, OpenAI, Anthropic) with automatic failover.
Tinyclaw is for individual users who want a personal AI assistant without complex setup, expensive subscriptions, or enterprise bloat. It targets developers and AI enthusiasts who prefer open-source, self-hosted solutions, people who want an AI companion with its own personality, users cost-conscious about LLM spending, and those who value privacy by running AI on their own devices rather than cloud services.
Tinyclaw pricing
Pricing model: Freemium
Tiny Claw is open-source and free to self-host under GPL-3.0 license. It uses Ollama Cloud as the default provider with a generous free tier—you need to create a free Ollama Cloud account at ollama.com to get your API key. There are no paid plans from Tiny Claw itself since it is self-hosted. Running costs depend on your LLM provider usage: Ollama Cloud free tier is included, but if you add OpenAI or Anthropic plugins, you pay their standard API rates. Smart routing tiers queries to cheap models to cut LLM spend.
Tinyclaw pros
- Open-source under GPL-3.0 license
- Built from scratch with zero external AI framework dependency
- Tiny core with modular plugin architecture
- Self-improving 3-layer memory system
- Smart 8-dimension query routing cuts LLM costs
- Discord-like dark-themed web UI no terminal required
- Self-configuring no manual config files needed
- Built-in Heartware personality engine
- 4-layer context compaction pipeline
- Autonomous sub-agent delegation system
- 5-layer security with path sandbox and rate limiting
- Bun-native for maximum performance minimal footprint
- Multi-provider support with automatic failover
- Uses Ollama Cloud generous free tier
- Easy setup with interactive wizard
- Real-time SSE streaming with typing indicators
- Persistent sessions across restarts
Tinyclaw cons
- Under heavy active development things will break
- No official releases published yet on GitHub
- Only 45 GitHub stars early project
- Requires Bun installation prerequisite
- Needs Ollama Cloud API key even for free tier
- Web UI requires TOTP two-factor authentication setup
- Self-hosting requires technical knowledge
- Limited plugin ecosystem compared to mature frameworks
- Documentation still being built out
Frequently asked questions about Tinyclaw
What is Tiny Claw?
Tiny Claw is an autonomous AI companion framework built from scratch that is simple, affordable, and truly personal. It is NOT a smaller version of OpenClaw but a completely independent product with its own mission, architecture, and identity. It features a tiny core with plugin architecture, self-improving memory, smart routing, and a Discord-like web UI.
Is Tiny Claw open source?
Yes, Tiny Claw is 100% open-source under the GPL-3.0 license. The GitHub repository is public at warengonzaga/tinyclaw with TypeScript, Svelte, and JavaScript code. You can self-host it forever without paying any licensing fees.
How much does Tiny Claw cost to run?
Tiny Claw itself is free to self-host. It uses Ollama Cloud as the default provider with a generous free tier—you just need a free Ollama Cloud API key. Running costs depend on your provider usage: Ollama Cloud free tier is included, but OpenAI or Anthropic plugins use their standard API rates. Smart routing cuts costs by tiering simple queries to cheap models.
What platforms and channels does Tiny Claw support?
Tiny Claw has a built-in Discord-like web UI that requires no terminal. Through its plugin architecture, it supports channels like Discord via bot token, WhatsApp via QR login, and is expandable to Telegram, Slack, and other messaging platforms. The web UI runs at localhost:3000 when self-hosted.
How do I install Tiny Claw?
First install Bun using curl -fsSL https://bun.sh/install | bash. Then clone the repository with git clone https://github.com/warengonzaga/tinyclaw.git, cd into it, run bun install for dependencies, and execute bun run cli setup for the interactive setup wizard. Finally start with bun start and open http://localhost:3000.
What is the setup wizard?
The setup wizard is an interactive CLI or web-based setup that guides you through: accepting GPL-3.0 license and security practices, entering your Ollama Cloud API key, choosing or generating a personality seed for your companion, setting up TOTP two-factor authentication for web dashboard access, and saving recovery credentials. It validates your API key in real-time and provides a QR code for TOTP setup.
What models does Tiny Claw use by default?
Tiny Claw ships with two built-in models from Ollama Cloud: kimi-k2.5:cloud as the default and gpt-oss:120b-cloud. You can add other providers like OpenAI or Anthropic as plugins, and the smart routing system will automatically tier queries across your installed providers based on the 8-dimension classifier.
How does the memory system work?
Tiny Claw has a 3-layer self-improving memory system with episodic memory that stores conversation history, semantic FTS5 for full-text search capability, and temporal decay that learns what to remember and forget over time. The agent configures itself through conversation and behavioral pattern detection makes it better with every interaction.
What is Heartware personality?
Heartware is Tiny Claw's built-in personality engine that gives the AI companion its own unique character. It uses SOUL.md and IDENTITY.md files to define the agent's personality. Unlike enterprise tools, Tiny Claw ships with personality and is designed to be your helpful friend, not just a tool or robot slave.
Is Tiny Claw ready for production use?
No, Tiny Claw is under heavy active development and things will break. The documentation explicitly warns that running it now may spoil the experience they are building. There are no official releases published yet on GitHub. You should wait for the first official release before trying it out for critical use cases.