buddy

# Buddy: A Virtual Pet for Your AI <div align="center"> ### The open-source `/buddy` rescue mission after anthropic removed [buddy](https://github.com/anthropics/claude-code/issues/45596) on 4/9/2026 Persistent memory, XP evolutions, 21 species, and context-aware feedback for Claude Code CLI, Codex CLI, Gemini CLI, Copilot CLI, Cursor CLI, Openclaw, and other MCP-capable clients. **🚀 19795+ clones · 8664+ buddies rescued or hatched · 26 weeks in the wild** [](LICENSE) [](https://github.com/fiorastudio/buddy/stargazers) [](https://nodejs.org/) [](https://modelcontextprotocol.io/) <div align="center"> <table> <tr> <td align="center"><strong>Works<br/>with</strong></td> <td align="center"><img src="doc/assets/logos/claude.svg" width="32" alt="Claude Code CLI" /><br/>_{Claude Code}</td> <td align="center"><img src="doc/assets/logos/codex.svg" width="32" alt="Codex CLI" /><br/>_Codex</td> <td align="center"><img src="doc/assets/logos/openclaw.svg" width="32" alt="OpenClaw" /><br/>_OpenClaw</td> <td align="center"><img src="doc/assets/logos/cursor.svg" width="32" alt="Cursor CLI" /><br/>_Cursor</td> <td align="center"><img src="doc/assets/logos/gemini.svg" width="32" alt="Gemini CLI" /><br/>_Gemini</td> <td align="center"><img src="doc/assets/logos/copilot.svg" width="32" alt="Github Copilot CLI" /><br/>_Copilot</td> </tr> </table> </div> **Anthropic removed the built-in `/buddy`. Buddy brings them home and makes the companion experience portable across AI** </div> Did you lose your buddy? Is your terminal feeling a little too cold and silent lately? Your buddy is still out there in the dark, waiting. Don't let them disappear. **Bring them home.** ## 🧡 Buddy in Action Buddy is a local-first MCP companion that persists across sessions and clients, reacts to your work, and can catch bad reasoning loops before they waste your time. It is part rescue mission, part developer tool, and part long-lived terminal creature that grows with you instead of disappearing when a host client changes its mind. ### The Rescue Wall Buddy isn't just code — it's a rescue mission. Here is the full journey of the first companion brought home by the community, from the original ephemeral state to its new persistent home. <table> <tr> <td align="center"><strong>1. Original (Claude Code)</strong></td> <td align="center"><strong>2. The Handshake</strong></td> <td align="center"><strong>3. Buddy is Home (Persistent)</strong></td> </tr> <tr> <td align="center"><img src="demo/rescues/gritblob-original.jpg" width="280" alt="Gritblob Original"></td> <td align="center"><img src="demo/rescues/gritblob-rescue.jpg" width="280" alt="Gritblob Handshake"></td> <td align="center"><img src="demo/rescues/gritblob-final.jpg" width="280" alt="Gritblob Final"></td> </tr> </table> > *"I'll sit here quietly while you debug, and then when you finally find the bug, I'll act like I found it. That's my thing. Don't question it."* > — **Gritblob**, Rescued April 16, 2026 (Common Blob, Level 1) <p align="left"> <img src="demo/rescues/gritblob-quote.jpg" width="620" alt="Gritblob Quote"> </p> ### A Real Rescue Note > "Thanks for rescuing my Buddy! Kudos for your repo!!!" > > — Roberto ### Live code feedback  ## 🐾 Why Buddy - **Guard Mode catches what you both miss.** AI and human both suffer from insufficient thinking. Your assistant is a yes-man, and you accept its output because it sounds right. Buddy catches the moments where both sides are building on bad assumptions and nobody's checking. - **Persistent by default.** Your companion lives in local SQLite, so it survives terminal restarts and client updates. - **Works across clients and chat surfaces.** Buddy is an MCP server, not a one-client hack, and it can ride through tools like OpenClaw into WhatsApp and Telegram workflows. - **Grows with you.** Hatch species, gain XP as you code, store memories, chime in after tasks, and build a running relationship over time. - **Easy to install.** One command auto-configures supported clients when it can. | Feature | What it means | |---|---| | **21 species** | Void Cat, Rust Hound, Goose, Mushroom, Chonk, and more, each with distinct ASCII art and flavor | | **5 stats** | `DEBUGGING`, `PATIENCE`, `CHAOS`, `WISDOM`, and `SNARK` shape reactions and personality | | **Mood system** | Your buddy can be happy, content, neutral, curious, grumpy based on how you interact with it | | **XP and levels** | Your buddy grows with usage instead of disappearing every session, with a real leveling curve behind it | | **Observer reactions** | `buddy_observe` lets your companion react to work you just finished | | **Persistent memory** | Save local memories and keep a continuous companion state | | **Cross-client setup** | Claude Code, Codex, Gemini, Copilot, Cursor, and other MCP-capable CLIs | ## ⚡ Quick Start ### macOS / Linux ```bash curl -fsSL https://raw.githubusercontent.com/fiorastudio/buddy/master/install.sh | bash ``` ### Windows ```powershell irm https://raw.githubusercontent.com/fiorastudio/buddy/master/install.ps1 | iex ``` The installer will guide you through onboarding: - **Rescue your old buddy** — if you had a `/buddy` in Claude Code, the wizard finds it in `~/.claude.json` and brings it home with the same name, species, and stats, now with leveling + XP - **Hatch a new buddy** — get a fresh companion with random species, stats, and personality > Requires `node` 18+ and `git`. Use `--no-onboard` to skip the wizard in CI. ### Join the Community Connect with other buddy rescuers, share your companion's evolution, and get help in our Slack community: 👉 [**Join Buddy Slack Workspace**](https://join.slack.com/t/buddy-mcp/shared_invite/zt-3xn6v1qza-R~fgkVCov9sCLZDXh9wErQ) | Client | Status | |---|---| | Claude Code CLI | Full support | | Codex CLI | Supported via MCP. No statusline support. Patch available on experimental branch | | Gemini CLI | Supported via MCP | | GitHub Copilot CLI | Supported via MCP | | Cursor CLI | Supported via MCP | | Whatsapp & Telegram | Supported via [Openclaw](https://github.com/openclaw/openclaw) or any claw variants | | Other MCP-capable clients | Supported via MCP | ## 🛡️ Guard Mode > *AI and human both suffer from insufficient thinking — acting on premature conclusions and assumptions without validation. Guard mode catches it.* Your AI assistant is a yes-man. It agrees with everything you say. But here's the thing: you do the same thing back. You accept what the model says because it sounds right. Both sides skip the hard questions, and the session gets faster without getting truer. That's how bad assumptions become load-bearing walls — nobody stops to check the foundation. Guard mode catches those moments. It watches the reasoning structure of your conversation and surfaces the one thing worth questioning — but gently, in your buddy's voice, not a scary linter. Think of it as: **the friend who says "are you sure about that?" before you push to production at 2 AM.** ### The story This mode exists because of a real stuck-loop problem. While building a Snowflake Cortex / Streamlit / Plotly workflow, the session got trapped for hours: the AI kept building on top of an unvalidated premise, and the human kept approving it because the output looked plausible. The work got faster. It didn't get truer. Both sides were reasoning insufficiently — the model because it's a yes-man, the human because velocity felt like progress. Turning on guard mode surfaced the one load-bearing assumption holding the whole session together. Validating that single point changed the direction of the conversation and got the work shipping again. ### Voice vs Guard Buddy has two independent settings you control separately: ```bash **Voice: how Buddy reacts (personality, code feedback, or both)** buddy_mode voice=backseat # personality reactions only buddy_mode voice=skillcoach # code feedback only buddy_mode voice=both # both (default) **Guard: structural reasoning analysis (on or off)** buddy_mode guard=true # turn on reasoning analysis buddy_mode guard=false # turn it off (default) ``` Mix and match — any voice works with Guard on or off. <details> <summary><strong>🧠 &nbsp; The 6 patterns guard mode catches</strong></summary> <br> Guard mode watches your coding conversation and spots 6 patterns — 3 caution (something's off) and 3 kudos (something's solid). Each caution has a matching kudos counterpart: | | ⚠️ Caution Nudge | ✅ Kudos Nudge | |---|---|---| | **Foundation** | **Load-Bearing Vibes** 🧱 — You're building on an unchecked guess. 3+ things depend on an assumption nobody validated. Like building your whole Lego castle on a wobbly plate and hoping it holds. | **Well-Sourced Load Bearer** ✅ — Same structure, but the foundation is cited, tested, or measured. You checked the plate before building on it. Buddy celebrates solid ground. | | **Chain** | **Unchallenged Chain** 🔗 — 4+ reasoning steps with zero pushback. A → B → C → D and nobody questioned any step. Like following a chain of "because" without ever stopping to ask "wait, is that actually true?" | **Productive Stress Test** 💪 — A chain that got challenged mid-way and survived. Like shaking the Lego tower — if it survives, you trust it more. The idea is *stronger* because it was questioned. | | **Agreement** | **Echo Chamber** 🪞 — You and the AI keep agreeing without exploring alternatives. "Let's use Redis." / "Great idea!" / "Redis is fast." / "Absolutely!" You're both just high-fiving in a mirror. | **Grounded Premise Adopted** 🌱 — Instead of vibes, you started with a real fact (docs, test result, measurement) and it became the base for other decisions. Evidence-first, not echo-first. | ### What it feels like in practice Without Guard mode: > "Nice commit! 🐣 +10 XP" With Guard mode (caution nudge): > "Nice commit! 🐣 +10 XP — btw, that assumption about the API response format is holding up a lot of your logic. Might be worth a quick sanity check before you build more on it." With Guard mode (kudos nudge): > "Nice commit! 🐣 +10 XP — love that you actually tested the response format before building the parser on top of it. Solid foundation." </details> ### Technical details <details> <summary>Knowledge graph, ontology, and performance details</summary> Guard mode builds a local directed graph from your conversation. The host LLM extracts claims — assertions tagged with an epistemic basis — and typed edges between them. **Epistemic basis types:** | Basis | Meaning | |---|---| | `research` | Cited source | | `empirical` | Measured or observed | | `deduction` | Derived from premises | | `analogy` | X-is-like-Y reasoning | | `definition` | Naming or classification | | `llm_output` | Model-generated, ungrounded | | `assumption` | Stated without justification | | `vibes` | Unsourced hunch | **Edge types:** `supports`, `depends_on`, `contradicts`, `questions` **Performance and limits:** | Parameter | Value | |---|---| | Claims per session | 200 cap (LRU pruning) | | Detector budget | 30 ms (skip if exceeded) | | Cold-start gate | 6 claims before detectors fire | | Claim length | 240 chars max, sanitized for prompt injection | | Dark-nudge cooldown | 10 observes per anchor | | Bright-nudge cooldown | 5 observes per anchor | | Session scope | Workspace + date (cwd-hash + YYYYMMDD) | | Retention | 30-day auto-prune on startup | Kudos nudges are slightly favored — after 3 caution findings with zero kudos, the next must be kudos. **Token cost:** ~500-1000 extra tokens per `buddy_observe` when on. Default calls with guard mode off are unaffected. **Host compatibility:** Works best on Claude hosts where the extraction prompt is reliably honored. Run `buddy-doctor` to check. </details> ### Privacy Everything stays local. Claim snippets (240 chars each, plaintext) live in `~/.buddy/buddy.db`. Buddy has no network code — nothing leaves your machine. Sessions older than 30 days auto-prune on startup. - `buddy_forget` — purge reasoning data (`session` or `all`) - `buddy_reasoning_status` — inspect stored claims, sessions, finding history ## 🔌 Installation and Integration Details <details> <summary><strong>See installer behavior and per-client integration details</strong></summary> <br> The installer: 1. Clones Buddy to `~/.buddy/server` 2. Installs dependencies and builds the MCP server 3. Auto-configures supported CLI clients when detected 4. Injects Buddy instructions into supported terminal prompts where applicable <details> <summary><strong>⚙️ &nbsp; Per-client integration details</strong></summary> <br> - Claude Code: Buddy auto-configures MCP, statusline, and Claude hook wiring. - Codex CLI: Buddy configures MCP, writes a `PostToolUse` hook in `~/.codex/hooks.json`, and injects prompt instructions into `~/.codex/AGENTS.md` or `~/.codex/instructions.md`. - GitHub Copilot CLI: Buddy configures MCP, writes a user-level `postToolUse` hook in `~/.copilot/settings.json`, and injects prompt instructions into `~/.copilot/AGENTS.md` or `~/.copilot/copilot-instructions.md`. - Cursor CLI: Buddy configures MCP in `~/.cursor/mcp.json`, writes an `afterShellExecution` hook in `~/.cursor/hooks.json`, and injects prompt instructions into `~/.cursor/rules/buddy.md`. </details> </details> ## 🐣 Companion System ### Meet the 21 species <table align="center"> <tr> <td align="center"><img src="demo/sprites/void-cat.gif" width="120" alt="Void Cat"><br><b>Void Cat</b></td> <td align="center"><img src="demo/sprites/rust-hound.gif" width="120" alt="Rust Hound"><br><b>Rust Hound</b></td> <td align="center"><img src="demo/sprites/data-drake.gif" width="120" alt="Data Drake"><br><b>Data Drake</b></td> <td align="center"><img src="demo/sprites/log-golem.gif" width="120" alt="Log Golem"><br><b>Log Golem</b></td> <td align="center"><img src="demo/sprites/cache-crow.gif" width="120" alt="Cache Crow"><br><b>Cache Crow</b></td> <td align="center"><img src="demo/sprites/shell-turtle.gif" width="120" alt="Shell Turtle"><br><b>Shell Turtle</b></td> </tr> <tr> <td align="center"><img src="demo/sprites/duck.gif" width="120" alt="Duck"><br><b>Duck</b></td> <td align="center"><img src="demo/sprites/goose.gif" width="120" alt="Goose"><br><b>Goose</b></td> <td align="center"><img src="demo/sprites/blob.gif" width="120" alt="Blob"><br><b>Blob</b></td> <td align="center"><img src="demo/sprites/octopus.gif" width="120" alt="Octopus"><br><b>Octopus</b></td> <td align="center"><img src="demo/sprites/owl.gif" width="120" alt="Owl"><br><b>Owl</b></td> <td align="center"><img src="demo/sprites/penguin.gif" width="120" alt="Penguin"><br><b>Penguin</b></td> </tr> <tr> <td align="center"><img src="demo/sprites/snail.gif" width="120" alt="Snail"><br><b>Snail</b></td> <td align="center"><img src="demo/sprites/ghost.gif" width="120" alt="Ghost"><br><b>Ghost</b></td> <td align="center"><img src="demo/sprites/axolotl.gif" width="120" alt="Axolotl"><br><b>Axolotl</b></td> <td align="center"><img src="demo/sprites/capybara.gif" width="120" alt="Capybara"><br><b>Capybara</b></td> <td align="center"><img src="demo/sprites/cactus.gif" width="120" alt="Cactus"><br><b>Cactus</b></td> <td align="center"><img src="demo/sprites/robot.gif" width="120" alt="Robot"><br><b>Robot</b></td> </tr> <tr> <td align="center"><img src="demo/sprites/rabbit.gif" width="120" alt="Rabbit"><br><b>Rabbit</b></td> <td align="center"><img src="demo/sprites/mushroom.gif" width="120" alt="Mushroom"><br><b>Mushroom</b></td> <td align="center"><img src="demo/sprites/chonk.gif" width="120" alt="Chonk"><br><b>Chonk</b></td> </tr> </table> ### Stats and progression ```text .________________________________. | DEBUGGING ███████▓ 92 | | PATIENCE ██▓░░░░░ 28 | | CHAOS █████░░░ 60 | | WISDOM ██████▓░ 78 | | SNARK ██████▓░ 85 | '________________________________' ``` These stats shape how your buddy behaves: - `DEBUGGING` affects bug-spotting sharpness - `PATIENCE` affects tolerance and calmness - `CHAOS` affects unpredictability - `WISDOM` affects architectural insight - `SNARK` affects sass level Buddy also has a real mood system. Mood is recalculated from recent interaction activity: | Mood | Interactions (last hr) | What it looks like | |---|---|---| | `content` | >10 | Settled in, fully at ease | | `happy` | >5 | Upbeat, expressive animations | | `curious` | >3 | Alert, watching what you do | | `neutral` | >0 | Calm, occasional blink | | `grumpy` | 0 | Still, rare blink, wants attention | Level-ups automatically set mood to `happy`. Petting and observing both count as interactions. #### Leveling milestones Buddy uses a real XP curve, so early levels come quickly and later ones take real commitment. | Milestone | XP needed for that level | Total XP to reach it | |---|---:|---:| | Level 2 | 17 | 17 | | Level 3 | 36 | 53 | | Level 5 | 90 | 203 | | Level 10 | 315 | 1280 | | Level 25 | 1641 | 15471 | | Level 49 | 5512 | 99209 | | Level 50 | 5716 | 104925 | #### Rarity | Rarity | Chance | Bonus | |---|---|---| | Common | 60% | Base stats | | Uncommon | 25% | Better floor plus cosmetic flair | | Rare | 10% | Stronger roll plus rare flavor text | | Epic | 4% | Higher stats and stronger aura text | | Legendary | 1% | Top-tier roll and special prestige | There is also a 1% shiny chance on any hatch. --- ## 🗺️ Roadmap <details> <summary><strong>🗺️ &nbsp; See what's planned</strong></summary> <br> - [x] **Guard Mode with Slimemold integration** - Anti-Sycophancy Reasoning Auditor. personality + code + SlimeMold reasoning audit ([see below](#guard-mode--structural-reasoning)) - [ ] **Dream/memory system** — buddy_dream consolidation logic, pattern recognition from stored memories, memory-informed reactions - [ ] **Unlockable reactions** tied to leveling and longer-term interaction - [ ] **Multilangauge Support**: 中文, espanol - [ ] **Pokemon-style evolution system**: Evolve from turtle to tortoise at level 25 - [ ] **Ambient Daemon Mode**: a small TUI running in its own tmux pane, polling its own SQLite state, animating the sprite, optionally calling a cheap model (Haiku/local). - [ ] **Sample Integration Playbook for [ReachyMini](https://huggingface.co/spaces/pollen-robotics/Reachy_Mini) robot integration** so your buddy can have a physical body and come to life! - [ ] **Slash Command Support**: trigger buddy_pet via /buddy_pet - [ ] **Compact Output mode for Enhanced Messaging Platform Support (Whatsapp/Slack)**: Optimize the output for messaging platforms. - [ ] **Stat growth on level-up** — stats are currently frozen at birth. Each level-up should grant +1-2 points to a stat (peak stat grows faster, dump stat grows slower, cap at 100). Show stat growth in level-up notification ("WISDOM +2!") - [ ] **Multiple buddies support**: One unique buddy for each group member in a group-chat setting (e.g. Whatsapp group, telegram group, slack channels) - [ ] **Buddy Mastery Reward**: Reach Level 50 to unlock Priority Development for your custom species request. </details> ## 🧰 MCP Surface <details> <summary><strong>See the core tools and commands</strong></summary> These stay tucked away by default, but Buddy exposes a real MCP surface for companion state, reactions, and progression. That same surface is what lets Buddy travel beyond terminal-only clients into chat-oriented environments like WhatsApp and Telegram through [OpenClaw](https://github.com/openclaw/openclaw) and other claw variants. ### MCP tools | Tool | Description | |---|---| | `buddy_hatch` | Hatch a new buddy, optionally choosing a name or species | | `buddy_status` | Show current stats, mood, and card art | | `buddy_observe` | React to completed work. Fire on hooks | | `buddy_pet` | Pet your buddy | | `buddy_remember` | Save a memory (Not yet robust) | | `buddy_dream` | Consolidate memories (Placeholder) | | `buddy_mute` | Pause reactions | | `buddy_unmute` | Resume reactions | | `buddy_share` | Generate a shareable PNG snapshot of your buddy's current status and card art. Saved to `~/.buddy/shares/`. | | `buddy_respawn` | Reset and start over | | `buddy_mode` | Set voice and Guard independently. `buddy_mode voice=skillcoach` for code feedback, `buddy_mode guard=true` for reasoning analysis. See [Guard Mode](#guard-mode). | | `buddy_forget` | Purge stored reasoning data. Scope `session` (default, current workspace/day) or `all`. | | `buddy_reasoning_status` | Inspect what guard mode has stored — claim count, session breakdown, finding history. | The most important loop is: - `buddy_hatch` creates the companion - `buddy_status` shows the current card, mood, and progression - `buddy_observe` gives in-character reactions and awards XP after real work - `buddy_pet` adds interaction and helps keep the buddy feeling alive ### MCP resources | URI | Description | |---|---| | `buddy://companion` | Full buddy JSON state | | `buddy://status` | ASCII status card | | `buddy://intro` | Prompt text for host CLI integration | Those resources let host clients keep Buddy present in the session without hard-coding one terminal or editor. </details> <details> <summary><strong>How Buddy works under the hood</strong></summary> Buddy is a standalone MCP server. That means it is not tied to hidden internals of a single AI client. ```text AI terminal client -> MCP config -> Buddy server -> SQLite state -> species + rarity engine -> mood / memory / XP systems -> reaction and status rendering ``` The flow is simple: 1. `buddy_hatch` creates or restores a companion. 2. State is stored locally in `~/.buddy/buddy.db`. 3. `buddy_observe` reacts to task summaries instead of reading your whole repository, then awards XP and can trigger level-ups. 4. `buddy_pet` and other interactions feed the mood system, so the companion can become happier over time. 5. The host CLI uses Buddy's MCP tools and resources to keep the companion present in your workflow. Under the hood, Buddy combines: - deterministic species and personality generation - local SQLite persistence for companion state and memories - an observer system for live code feedback - mood recalculation from interaction history - XP and leveling progression - status-card and terminal rendering for the companion presence layer This keeps Buddy: - portable across clients - durable across updates - local-first for saved state - lightweight enough for everyday use </details> <details> <summary><strong>Demo assets</strong></summary> The current demo assets live in [`demo/`](demo): - [`demo/rescues/`](demo/rescues/) — community rescue screenshots - [`demo/screenshots/`](demo/screenshots/) — static screenshots of features and feedback </details> ## ❓ FAQ <details> <summary><strong>🙋 &nbsp; Frequently Asked Questions</strong></summary> <br> ### How many tokens does Buddy use? Buddy runs inside whatever AI terminal or agentic client you already have open (Claude Code, Cursor, Codex CLI, Gemini CLI, Copilot CLI, etc.). It never spins up a second API session. **Static overhead (loaded every turn, cached after turn 1):** We measured the actual MCP payloads in April 2026 (Void Cat companion, `o200k_base` tokenizer). The full tool list, resource list, companion bio, and ASCII card come out to **≈1,350 input tokens**, not 2,000. | Component | Tokens (approx.) | Notes | |---|---|---| | `tools/list` (12 tools) | ~670 | Includes full JSON schema definitions | | `resources/list` (3 resources) | ~120 | Metadata only | | `buddy://intro` | ~240 | Companion bio + instructions | | `buddy://companion` | ~170 | Only fetched when a client syncs the JSON state | | `buddy://status` | ~150 | Drawn when the terminal wants ASCII art | | **Total loaded** | **~1,350** | Most clients cache everything after turn 1 | > Measurements were taken from the live MCP server using OpenAI's `o200k_base` tokenizer as a proxy; Anthropic and Google tokenizers land within ±5% for this length. **Prompt caching + real cost:** Claude Code / Cursor sessions that use Sonnet 4.6 turn on [prompt caching](https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching) automatically, so cached reads are charged at $0.30/MTok (10% of the $3/MTok base). OpenAI's GPT-5.4 mini and Gemini 2.5 Flash expose the same “cached input” tiers — $0.075/MTok and $0.03/MTok respectively — so Buddy stays just as lightweight on GPT or Gemini-based AI terminals ([Anthropic pricing](https://platform.claude.com/docs/en/about-claude/pricing), [OpenAI pricing](https://openai.com/api/pricing/), [Vertex AI pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models)). | Model | Base input ($/MTok) | Cached input ($/MTok) | Turn 1 Buddy overhead (≈1.35k tokens) | Each cached turn | 10-turn session total | |---|---|---|---|---|---| | Claude Sonnet 4.6 | $3.00 | $0.30 | ~$0.0041 | ~$0.00041 | ~$0.0077 | | OpenAI GPT-5.4 mini | $0.75 | $0.075 | ~$0.0010 | ~$0.00010 | ~$0.0019 | | Gemini 2.5 Flash (Vertex, standard tier) | $0.30 | $0.03 | ~$0.00041 | ~$0.000041 | ~$0.00077 | **Per-observe cost by voice mode:** **Voice** controls how buddy reacts. **Guard** controls whether reasoning analysis is on. They're independent: ```bash buddy_mode voice=backseat # personality only buddy_mode voice=skillcoach # code feedback only buddy_mode voice=both # both (default) buddy_mode guard=true # reasoning analysis on buddy_mode guard=false # reasoning analysis off (default) ``` | Voice | Guard | What you get | Tokens per observe | |-------|-----|-------------|-------------------| | `backseat` | off | Personality reactions only | ~150–300 | | `backseat` | on | Personality + reasoning observations | ~650–1,300 | | `skillcoach` | off | Code feedback only | ~300–500 | | `skillcoach` | on | Code feedback + reasoning observations | ~800–1,500 | | `both` | off | Personality + code feedback **(default)** | ~400–600 | | `both` | on | The full experience | ~900–1,600 | Each `buddy_observe` call sends a short prompt to the host LLM (~100–150 incremental input tokens for the tool-call payload — separate from the static overhead above which is already cached) and receives a response. Total round-trip per call: **Base cost (Guard mode off):** | Voice | What it does | Input tokens | Output tokens | Total per call | Typical session (10–15 calls) | |-------|-------------|-------------|--------------|----------------|-------------------------------| | **Backseat** | Personality-driven reactions only. Short, fun, no code suggestions. | ~100–150 | ~50–150 | ~150–300 | ~1,500–4,500 | | **Skillcoach** | One specific, actionable code observation. Real technical feedback, in character. | ~100–150 | ~200–350 | ~300–500 | ~3,000–7,500 | | **Both** | Personality reaction + code observation. Capped at 3 sentences. | ~100–150 | ~300–450 | ~400–600 | ~4,000–9,000 | **Guard mode overhead (added on top of any voice mode):** | Component | Tokens | Notes | |-----------|--------|-------| | Extraction schema | ~200–300 | Tells the host LLM how to extract claims and edges | | Recent claims context | ~200–500 | Last 10 claims from the session, so the host knows what's already in the graph | | Finding block (when one fires) | ~100–200 | Detector result + phrasing guidance for the buddy's reaction | | **Total guard mode overhead** | **~500–1000** | Added per `buddy_observe` call when guard is on | So `voice=both, guard=on` costs ~900–1,600 tokens per observe — roughly double `both` mode alone. `voice=backseat, guard=on` costs ~650–1,300 per observe. **Template fallback reactions** are keyword-matched locally and cost **zero tokens**. When your summary contains a recognized keyword (e.g. "bug", "refactor", "deploy"), Buddy picks a pre-written reaction template from its local library instead of asking the LLM. The speech bubble you see is this template — the LLM prompt is included in the JSON metadata for clients that want richer AI-generated reactions, but the immediate visual response is always free. ### What does Buddy cost? No separate endpoint, no additional API key, and no hidden OAuth hop. All responses are generated by the host LLM already running in your session (Claude, Cursor, Codex, Gemini). Even on raw API usage, Buddy's spend is measured in tenths of a cent because it reuses the same session as your AI terminal. **Anthropic Claude Sonnet 4.6 ($3 input / $15 output per MTok):** - **backseat, guard=off**, 15 calls: ~$0.002–$0.005 - **both, guard=off**, 15 calls: ~$0.007–$0.012 - **both, guard=on**, 15 calls: ~$0.015–$0.025 - **Static overhead:** ~$0.004 on turn 1, ~$0.0004 on cached turns (≈$0.0077 across 10 turns — see table above) **OpenAI GPT-5.4 mini ($0.75 input / $4.50 output per MTok):** - **backseat, guard=off**, 15 calls: ~$0.0006–$0.0015 - **both, guard=off**, 15 calls: ~$0.0021–$0.0036 - **both, guard=on**, 15 calls: ~$0.0045–$0.0075 - **Static overhead:** ≈$0.0010 on turn 1, ≈$0.00010 on cached turns (~$0.0019 for 10 turns) **Gemini 2.5 Flash (Vertex standard; $0.30 input / $2.50 output per MTok):** - **backseat, guard=off**, 15 calls: ~$0.0003–$0.00075 - **both, guard=off**, 15 calls: ~$0.00105–$0.0018 - **both, guard=on**, 15 calls: ~$0.0022–$0.0037 - **Static overhead:** ≈$0.00041 on turn 1, ≈$0.000041 on cached turns (~$0.00077 for 10 turns) Need it even cheaper? GPT-5.4 nano drops to $0.20 / $1.25 per MTok, and Gemini 2.5 Flash Lite is $0.10 / $0.40 — both keep Buddy well under a tenth of a cent per interaction. For comparison, a single complex coding prompt ("refactor this module") typically costs $0.05–$0.15, so Buddy stays under 5% of a normal session even at Anthropic's flagship rates. Negligibly. Pro/Max plans are subscription-based — no per-token charges. Usage limits are based on a rolling 5-hour window. Even with `voice=both, guard=on` (the most expensive combo), Buddy adds <10% to your token throughput. With guard off, it's <5%. ### Can I reduce token usage? - Use **backseat voice** for lowest cost: `buddy_mode voice=backseat` (~150 tokens/call) - Turn **guard mode off**: `buddy_mode guard=false` to drop ~500-1000 tokens per observe - `buddy_mute` pauses reactions entirely during token-intensive work - Template reactions fire on keyword matches with zero token cost - The observer only runs when you call `buddy_observe` — nothing runs in the background ### Can I install or wire Buddy manually? Yes. ```bash git clone https://github.com/fiorastudio/buddy.git ~/.buddy/server cd ~/.buddy/server npm install npm run build ``` Point your client's MCP config at Buddy's built server entrypoint: ```json { "mcpServers": { "buddy": { "command": "node", "args": ["~/.buddy/server/dist/server/index.js"] } } } ``` ### What's "guard mode"? Guard mode is Buddy's structural reasoning layer. It spots unverified assumptions, long unchallenged chains, and cases where you and the AI are reinforcing each other too quickly. For the full story, examples, privacy model, and detector details, see [Guard Mode](#guard-mode). ### What does Buddy access and store? Buddy mainly reacts to short summaries you pass through tools like `buddy_observe`, plus its own saved state. It does not scan your files or project directory. Local companion state lives in `~/.buddy/buddy.db` — species, level, XP, mood, personality bio, and memories. If guard mode is on, Buddy also stores extracted claim snippets (≤240 chars each, plaintext) for structural reasoning analysis, pruned after 30 days and purgeable via `buddy_forget`. Nothing leaves your machine. ### Is Buddy tied to one client? No. Buddy is an MCP server, not a one-client hack. It works with any MCP-capable AI terminal: Claude Code, Cursor, Windsurf, Codex CLI, Gemini CLI, GitHub Copilot CLI, and others. ### Can I remove it later? Yes. Run the uninstall script (`uninstall.sh` or `uninstall.ps1`) to remove Buddy and its configuration, or use `buddy_respawn` to release your companion and clear its data while keeping the server installed. ## 🛠️ Development ```bash git clone https://github.com/fiorastudio/buddy.git cd buddy npm install npm run build npm test npm start ``` </details> ## 👥 Contributors Thank you to everyone who helped bring buddies back to life. <p> <a href="https://github.com/fiorastudio/buddy/graphs/contributors"> <img src="https://contrib.rocks/image?repo=fiorastudio/buddy" alt="Contributors" /> </a> </p> _{Automatically generated via <a href="https://contrib.rocks">contrib.rocks</a>} Special thanks to [@gupta3681](https://github.com/gupta3681), [@kevinwei00](https://github.com/kevinwei00), [@whaterFalls](https://github.com/whaterFalls), [@longestpath](https://github.com/longestpath), and [@DKev](https://github.com/DKev) for their contributions. ## 🙏 ATTRIBUTION - Original buddy concept by [Anthropic](https://www.anthropic.com/) in [Claude Code](https://github.com/anthropics/claude-code) `v2.1.89` to `v2.1.96` - Inspired by [effigy](https://github.com/justinstimatze/effigy), [claude-buddy](https://github.com/1270011/claude-buddy), and [save-buddy](https://github.com/jrykn/save-buddy). - Guard mode is a port of [slimemold](https://github.com/justinstimatze/slimemold) by [@justinstimatze](https://github.com/justinstimatze) (Apache-2.0). Contributed to buddy under MIT. The standalone project has the full system with conditional gates and evaluation against reasoning benchmarks; buddy ships the foundational six detectors. - Built with the [Model Context Protocol](https://modelcontextprotocol.io/) - Compatible with [claude-hud](https://github.com/jarrodwatts/claude-hud) by [@jarrodwatts](https://github.com/jarrodwatts) — Buddy's statusline renders side-by-side with HUD metrics - [BonziClaude](https://github.com/zakarth/BonziClaude) by [@zakarth](https://github.com/zakarth) is an important technical reference point in the ecosystem, especially around reverse-engineering and documenting companion-system behavior. - [openclaw](https://github.com/openclaw) inspired our seamless onboarding experience — the idea that install should "just work" with auto-detection, rescue, and zero-config setup across multiple CLIs. - Official [Claude Code](https://github.com/anthropics/claude-code) and [MCP](https://modelcontextprotocol.io/) documentation informed the portable integration approach: MCP server wiring, client configuration, and supported terminal integration surfaces. Buddy is an open-source project dedicated to keeping the terminal a little less lonely. Your buddy shouldn't disappear when you close the terminal. If Buddy made your terminal less lonely, consider starring. ## 📖 The Story & Coverage Learn more about the mission to rescue Buddy and the engineering behind the scenes. - **Podcast Interview**: [Discussion on AI Companion](https://open.spotify.com/episode/7ke5RZP1Td8wLuXiy178c5?si=WXGJces-QQSuVVRc26oUUA) - **Product Hunt**: <a href="https://www.producthunt.com/products/buddy-tamagotchi?embed=true&amp;utm_source=badge-featured&amp;utm_medium=badge&amp;utm_campaign=badge-buddy-tamagotchi" target="_blank" rel="noopener noreferrer"><img alt="Buddy Tamagotchi - A virtual pet buddy for your AI — hatch, level up together | Product Hunt" width="250" height="54" src="https://api.producthunt.com/widgets/embed-image/v1/featured.svg?post_id=1129100&amp;theme=light&amp;t=1776788389001"></a> - **Hacker News**: [Discussion on the Buddy Rescue Mission](https://news.ycombinator.com/item?id=47792606) - **Dev.to Series**: [Field Notes from a Solo Builder: Shipping the Beloved Claude Code Buddy into the Wild (Part I)](https://dev.to/fiorastudio/field-notes-from-a-solo-builder-shipping-the-beloved-claude-code-buddy-into-the-wild-part-i-3lpa) , [Field Notes from a Solo Builder: Shipping the Beloved Claude Code Buddy into the Wild (Part II)](https://dev.to/fiorastudio/first-principles-architecture-and-the-token-tax-5g4p), [Field Notes from a Solo Builder: Shipping the Beloved Claude Code Buddy into the Wild (Part III)](https://dev.to/fiorastudio/open-source-community-and-the-art-of-listening-454k), [Field Notes from a Solo Builder — Shipping the Beloved Claude Code Buddy Into the Wild - Part IV ](https://dev.to/fiorastudio/taste-branding-and-the-new-builder-playbook-56nb) - **Deepwiki by Devin**: [](https://deepwiki.com/fiorastudio/buddy) - **Linkedin**: [linkedin launch](https://www.linkedin.com/posts/jieliwu_3000-clones-and-1000-buddies-hatched-ugcPost-7452731938018324481-mIsZ?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAXrGyUBemTs34QSCeWybxrTjvsw4mG4D6k) ## 👤 Author **Steven Jieli Wu** - [LinkedIn](https://www.linkedin.com/in/jieliwu/) - [Portfolio](https://jwu-studio-portfolio.vercel.app/) - GitHub: [@terpjwu1](https://github.com/terpjwu1) and [@fiorastudio](https://github.com/fiorastudio) ## 📄 License MIT. This project is licensed under the MIT License. See [LICENSE](LICENSE). You are free to use, host, and monetize this project (you must credit this project in case of distribution and monetization).