ShibaClaw

ShibaClaw is an open-source, self-hosted AI agent that runs on your own machine or server, published by RikyZ90 under the Apache-2.0 license. It combines a Python agent engine with a browser WebUI, a terminal REPL, and a Windows desktop app. The latest release listed is v1.1.0 (2026-10-18).

What It Is

ShibaClaw is a modular personal-assistant agent framework. A central gateway routes messages from channels such as the WebUI, Telegram, Discord, and Slack through an asynchronous message bus to the agent loop (ShibaBrain), which calls LLM providers and registered tools. It uses native provider SDKs for 28 providers rather than a LiteLLM proxy layer, and also supports local endpoints such as Ollama, LM Studio, llama.cpp, and vLLM.

Security Model

The README says defenses ship in the core engine rather than as external glue: install-time CVE auditing of pip and npm, prompt-injection wrapping of tool results in randomized boundaries, shell hardening with deny patterns, SSRF and DNS-rebinding protection, a workspace sandbox, an encrypted credential vault, and bearer-token access control.

Memory, Knowledge and Automation

Memory has three tiers: per-session working memory, cross-session semantic memory backed by FAISS, and procedural memory saved as skills and schedules. An optional local RAG knowledge base indexes PDF, CSV, HTML, TXT, and Markdown into named collections that the agent can search. Background automations run on cron-like schedules or event triggers in isolated sessions, and an opt-in self-evolution feature uses validation gates.

Integrations and Setup

ShibaClaw connects to MCP servers, auto-registers their tools, and integrates with Klavis for one-click OAuth to SaaS apps. Channel and TTS plugins are discovered through Python entry points, including a WhatsApp Web bridge and an offline Supertonic speech engine. Installation options are a Windows PowerShell installer or setup executable, a Linux/macOS install script, Docker Compose, or pip (Python 3.12+).

Update: v1.1.0

The v1.1.0 notes describe background tasks that keep status across chat reconnects, separate execution from optional project verification, show exit codes and bounded logs, and improved chat rendering and Windows installation with SHA256 verification.