OctopusGarage - Build More. Reach Everywhere.
Compact open-source tools for durable AI-assisted work.
OctopusGarage builds the operating layer around long-running agents: remote control, local machine upkeep, source capture, private memory, learning loops, practical judgment, local communication, and focused browser products.
Website Β· Project index Β· llms.txt Β· GitHub
Agent control Private memory Local automation GitHub Pages
| Use case | Project | Why it exists |
|---|---|---|
| Keep coding agents reachable | tmux-claude-bot | Control Claude Code and OpenAI Codex sessions from Telegram, Feishu/Lark, CLI, or TUI while the real process stays inside tmux. |
| Bridge one chat to one pane | telegram-bridge | Forward Telegram messages into tmux and stream pane output back to chat. |
| Preserve working memory | alcove | Keep personal memory local-first for agent workflows and private retrieval. |
| Capture source material | clipsmith | Save posts, articles, media, and OCR into portable local bundles. |
| Learn inside the video | fluent-frame | Turn YouTube into a local-first bilingual English-learning workspace. |
| Apply judgment frameworks | phronesis | Use project-local agent modules for practical philosophy and reasoning lenses. |
| Keep repositories moving | git-auto-sync | Monitor local repos, stage carefully, write conventional commits, push, and notify. |
| Stabilize the machine | net-auto-switch | Measure WiFi and Clash Verge proxy quality, then switch only when the signal justifies it. |
Agent control
- tmux-claude-bot - multi-project agent sessions in tmux, controlled from chat, CLI, or TUI.
- telegram-bridge - focused Telegram-to-tmux bridge with streamed pane output.
Agent skills
- octopusgarage-skills - reusable Claude Code and Codex skills for OctopusGarage workflows.
Local automation
- git-auto-sync - AI-assisted multi-repo sync with scheduler support and pluggable notifications.
- net-auto-switch - layered WiFi and proxy-node switching for macOS and Clash Verge.
- envsentinel - terminal-first audit of local environment signals for AI coding CLIs.
Knowledge and learning
- alcove - local-first personal memory for agent workflows.
- clipsmith - source capture for articles, posts, media, and OCR.
- english-pilot - conversation gate, coaching assistant, and English review system for AI-assisted work.
- fluent-frame - local-first Chrome extension for bilingual YouTube subtitles and English learning notes.
- phronesis - practical-wisdom workspace for applying philosophical thought modules with local agents.
Private communication and browser labs
- mesh-talk - serverless LAN messaging with a Rust core and Tauri desktop app.
- bracket-guess - browser-native tournament bracket picker with poster export.
- MBTI Lab - bilingual personality test with shareable results and offline-friendly behavior.
| Principle | What it means here |
|---|---|
| Local first | Private data and working state should stay on the machine whenever possible. |
| Operator visible | Automation should explain what it is doing and leave room for approval. |
| One-command install | Tools should be easy to try, update, and remove. |
| Small surface area | Each repository should solve one concrete workflow well. |
| Useful before impressive | Polish matters, but the first job is to save time or reduce friction. |
| Surface | Link |
|---|---|
| Organization site | https://octopusgarage.github.io/ |
| Project data | https://octopusgarage.github.io/projects.json |
| AI-readable summary | https://octopusgarage.github.io/llms.txt |
| Full AI profile | https://octopusgarage.github.io/llms-full.txt |
Build More. Reach Everywhere.