Index / BashoBot · updated Aug 17, 2026
BashoBot
uraimo/bashobot · healthy · rank 43 of 43 by stars
A delightfully audacious OpenClaw clone written entirely in ~4k lines of Bash. It proves you don't need Node.js or a heavy runtime—just pipes, jq, and sheer determination.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 55 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 20 files examined
3 providers · custom endpoint
Pinned models
gpt-4o released 2024-05-13 — 2 years old (from the public model catalogue)
Pin last edited 6 months ago
Evidence
Decision
Why choose BashoBot over OpenClaw?
Why choose this
- Zero runtime dependencies beyond bash 3.2+ and standard Unix tools
- Extremely low resource footprint (sub-2MB memory, near-instant boot)
- Trivially containerizable with a single-file daemon
Tradeoffs
- No MCP (Model Context Protocol) support or structured plugin system
- Shell-based exec tool is inherently risky even with whitelisting
- No multi-channel support beyond Telegram and CLI
Best fit
- Developers who want an AI assistant on ultra-minimal systems (no Node.js/Python)
- Containerized or embedded deployments where every megabyte counts
- Bash enthusiasts who want to hack on and extend their assistant with shell scripts
Avoid if
- You need production-grade security sandboxing (shell access risk is inherently high)
- You rely on a rich plugin ecosystem or MCP server support
- You need multi-user support or enterprise auth flows
High confidence in technical assessment based on README, commits, and release data. Community sentiment is inferred from zero Reddit mentions and limited web presence, indicating very early-stage adoption.
AI layer reviewed Aug 2, 2026 · how this is written
Star activity
6 stars today
Overview
BashoBot is a personal AI assistant implemented entirely in Bash (compatible with version 3.2+), inspired by OpenClaw's architecture but deliberately stripped of all runtime dependencies beyond standard Unix utilities. The project uses a modular named-pipe architecture where a central daemon process coordinates message routing between a Telegram bot interface, a CLI client, and various AI provider backends (OpenAI, Anthropic, and subscription-based providers via OAuth).
At its core, BashoBot replicates the essential OpenClaw feature set: tool execution with an optional whitelist/approval system, markdown-based persistent memory, shared session management, and a SOUL.md personality file. The tool system is particularly notable—it allows the LLM to execute arbitrary shell commands, which is both powerful and dangerous. The project mitigates this with a configurable whitelist and user-approval flow, but the inherent risk of shell access in a Bash environment remains high. Recent commits show active maintenance with fixes for tool-calling behavior (configurable timeouts, preventing excessive read_file calls) and provider configuration churn.
The architecture is refreshingly transparent: the entire codebase is approximately 4,000 lines of Bash that can be read and audited in an afternoon. This makes it an excellent learning resource for understanding how AI agent harnesses work under the hood. However, the trade-offs are significant—no MCP support, no plugin ecosystem, no voice capabilities, and a shell execution model that would make any security-conscious team nervous. For the right use case (containerized, single-user, low-stakes environments), BashoBot is a clever and functional alternative to heavier agent frameworks.