by openwalrus
The composable agent runtime.
# Add to your Claude Code skills
git clone https://github.com/openwalrus/walrusThe composable agent runtime. Compact daemon core. Memory, channels, tools — all hooks. Use what you need, skip what you don't.
curl -fsSL https://openwalrus.xyz/install.sh | sh
Or cargo install openwalrus. See the installation guide for details.
# Start the daemon
walrus daemon
# Chat with your agent
walrus attach
Point it at any LLM — Ollama, OpenAI, Anthropic, DeepSeek, or any OpenAI-compatible API.
[walrus]
model = "qwen3:4b"
[model.qwen3]
base_url = "http://localhost:11434/v1"
Full config reference: configuration.
Walrus is a daemon that runs agents and dispatches tools. The daemon ships with built-in tools (file I/O, shell, task delegation), MCP server integration, and [skills] (Markdown prompt files).
Heavier capabilities live outside the daemon as WHS services — managed child processes you add or remove in config:
<!-- badges --> <!-- docs -->No comments yet. Be the first to share your thoughts!
| Service | What it does | | ------------------ | -------------------------------------------- | | Memory | Graph memory — LanceDB + semantic embeddings | | Search | Meta-search aggregator | | Gateway | Telegram, Discord adapters |
The daemon stays small. Services scale independently.
GPL-3.0