henrykbrzoska

Bebok — AI skill for Claude Code

AI community

Desktop GUI Headless Local-first AI coding agent in Rust (axum, tokio, rmcp, portable-pty) behind HTTP + SSE, with an Angular 20 zoneless client, Tauri 2 desktop shell(and mobile in near future).

How to install Bebok

This entry records only its repository, not the path inside it, so there is no exact command to give. Open henrykbrzoska/bebok and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Bebok does

Desktop GUI Headless Local-first AI coding agent in Rust (axum, tokio, rmcp, portable-pty) behind HTTP + SSE, with an Angular 20 zoneless client, Tauri 2 desktop shell(and mobile in near future). OpenAI/Anthropic/Z.ai/Ollama + more, MCP bridge, permission engine, PTY terminal.

Alternatives in AI

Dive Into Claude Code — A Systematic Analysis and Discussion of Claude Code for Designing Today's and Future AI Agent Systems 2.1k ★
HolyClaude — AI coding workstation: Claude Code + web UI + 7 AI CLIs + headless browser + 50+ tools 2k ★
Open Science — Open Science Desktop — local-first, model-agnostic AI research workbench for macOS, Windows & Linux 1.5k ★

README

Bebok 1.5.0

Bebok is a local-first AI coding agent. A headless Rust engine (axum + tokio, HTTP + SSE + WebSocket) owns sessions, tools, permissions, sub-agents and configuration; a thin Angular client renders it. The client runs as a desktop app (Tauri 2, engine bundled as a sidecar) or in the browser against an engine you start yourself. Everything runs on your machine; the only network traffic is to the LLM providers you configure.

Roadmap: 1.6.0 adds an Android app (chat + remote follow).

![Chat with collapsed tool groups, context meter and the right drawer](docs/screenshots/chat-tool-groups.png)

![Agents panel with live sub-agent tasks](docs/screenshots/agents-panel.png)

![Changes panel with a unified diff of a file modified by the agent](docs/screenshots/changes-diff.png)

![Explorer with syntax highlighting](docs/screenshots/explorer-highlight.png)

![Provider settings with a masked API key](docs/screenshots/settings-providers.jpg)

Features

Chat

Every tool call collapses into a one-line row; consecutive tool-only turns merge into one group with summed In/Out usage (group -> call -> arguments/output expand).
Safety-level dots on each tool call (safe / caution / dangerous / uncategorized) with a legend.
Context meter (tokens sent vs. the model window from the catalog), "Compact context" button, automatic compaction at 85 % with a server-side marker.
Typed inline chips for paths, commands and identifiers; file paths and .md links open the Preview panel; bare URLs are clickable.
Syntax highlighting (highlight.js, lazy per language) shared by chat, Preview, Explorer and the diff view; ATX headings; model text is HTML-escaped.
Long transcripts render the last 60 messages with "Load earlier messages".
Image attachments (png/jpeg/webp/gif, 5 per prompt, 5 MiB each) for vision models.
Thinking effort per session (off / low / medium / high / max).