VibeZoo — Intelligent Companion Extension for AI Coding Assistants

VibeZoo = Crow Memory (Synaptic Memory) + VibeZoo MCP Bridge (33 Tools)

VibeZoo is a Companion Extension for Zoo Code. Without modifying a single line of Zoo Code's source code, it enables the LLM to search, analyze, review, and document code more intelligently. It remembers your habits and preferences, and enables real-time visual collaboration (Whiteboard, Dropzone, Vision AI).

---

Support VibeZoo Development ☕

If you find VibeZoo and Crow Memory helpful for your productivity, consider supporting our development! 💖 Sponsor VibeZoo on Gumroad

Your support helps us develop new features and keep the AI coding revolution moving forward.

---

🌍 Global i18n / l10n Support

VibeZoo fully supports global internationalization (i18n) across 20 languages (English base + ar, bg, cs, de, es, fr, he, hu, it, ja, ko, pl, pt-BR, ru, th, tr, vi, zh-CN, zh-TW). Built on top of the native vscode.l10n API, VibeZoo automatically detects your VS Code display language, ensuring a native experience regardless of your region.

---

🚀 Out-of-the-Box Setup (Universal UX)

Getting started with VibeZoo is easier than ever. We provide a one-click bootstrapper to set up the Python environment, install dependencies, and build the frontend extension all at once.

Standard Directory Layout

VibeZoo now follows the same standard path scheme as Crow Memory:

Platform	Runtime Directory
Windows	%USERPROFILE%\mcp-servers\vibezoo\
macOS/Linux	~/mcp-servers/vibezoo/

The auto-start command in MCP settings is:

cd /d "%USERPROFILE%\mcp-servers\vibezoo" && start_vibezoo_bridge.bat

All MCP servers (crow-memory, vibezoo, etc.) coexist under %USERPROFILE%\mcp-servers\.

1. Clone the repository:

   git clone https://github.com/vibezoo/VibeZoo_forZoocode.git
   cd VibeZoo_forZoocode

2. Run the bootstrapper:

   • Windows: Double-click init_vibezoo.bat or run it in the terminal.
   • macOS/Linux: Run bash init_vibezoo.sh.

   This creates the standard runtime directory, copies bridge files, sets up a Python venv, and builds the extension.

3. Auto-Connect: Once the VibeZoo extension is active, it automatically starts the Python MCP Bridge on port 9027, resolves a working Python interpreter across python/python3/venv environments, and keeps the global mcp_settings.json synchronized so Zoo Code connects via Streamable HTTP automatically.

4. Global Mode Installation (via vibezoo_setup — Recommended): VibeZoo provides a one-command setup that automatically installs all 6 custom modes (orchestrator-crow, project-research, architect, code, debug, ask) with VibeZoo tool priority enabled across ALL modes.

   Step 1: Start the VibeZoo Bridge:

   python mcp-servers/vibezoo_mcp_bridge.py --port 9027

   Step 2: In your Zoo Code chat, run the setup tool:

   vibezoo_setup(target="minimal", configure_custom_modes=True)
   

   This will:

   • Install/verify Python dependencies (fastmcp, uvicorn, starlette)
   • Configure global mcp_settings.json with VibeZoo Streamable HTTP endpoint
   • Configure .zoo/config.json
   • Install 6 custom modes to Zoo Code's global custom_modes.yaml with VibeZoo tool priority enabled

   No regex errors! All modes are instructed to prefer VibeZoo MCP tools (search_codebase, review_code, review_project, etc.) over native tools. VibeZoo tools handle invalid regex gracefully with automatic substring fallback.

   Tip: You can still use the manual template from global_install_templates/vibezoo_mode.yaml if you prefer a hands-on approach. The vibezoo_setup tool automates this process for convenience.

---

1. VibeZoo MCP Bridge — Tool Overview (33 Tools)

The VibeZoo MCP Bridge operates based on FastMCP + Streamable HTTP, communicating with the Zoo Code MCP client via vibezoo_mcp_bridge.py at localhost:9027/mcp. It provides a total of 33 MCP tools through a modular architecture (bridge/tools/).

Key infrastructure modules:

• ast_singleton.py — Shared AST engine singleton (consolidated from 5 duplicated copies)
• fuzzy_matcher.py — Trigram Dice coefficient fuzzy matching engine
• embedding_client.py — Embedding-based semantic search client (Ollama/OpenAI auto-detection)
• result_ranker.py — BM25 + embedding cosine similarity result ranking
• file_cache.py — L1 memory cache with 20s TTL for search results

1.0 Autonomous Agents — Feedback & Telemetry

The vibezoo_feedback tool lets the LLM write telemetry logs (feedbacks/) to suggest new capabilities or highlight repetitive tasks for continuous improvement. (Web search tools are covered in 1.11 Web.)

1.1 UX (2 Tools) — Intent Detection + Auto Tool Chains

When you say "I'll show you a file", the Dropzone opens. Uploaded files are automatically analyzed through the SSA→OCR→MiniCPM pipeline.

Tools: ux_coordinator, analyze_uploaded_file

The ux_coordinator tool is now Crow Memory-aware — when keyword-based intent detection has low confidence, it queries Crow Memory for recent context (dropzone uploads, conversation history) to disambiguate. Detected intents include file_share, drawing_request, code_analysis, project_setup, and the new fix_loop — automatic bug fix / error recovery.

1.2 Scout (3 Tools) — Code Search and Exploration

Tools: search_codebase, find_references, summarize_architecture

Quickly grasp the project structure and find symbols or functions accurately using tree-sitter AST. target_path parameter added: Enables global search in a specific directory (e.g., search_codebase(query=..., target_path="C:/Projects/MyApp")).

Search modes (v0.16.0 overhaul):

• auto (default) — AST-aware search with automatic fallback
• exact — Literal substring matching
• fuzzy — Real trigram Dice coefficient approximate matching via fuzzy_matcher.py (previously identical to auto; now performs genuine fuzzy matching)
• ast — tree-sitter AST symbol extraction
• semantic — Embedding-based cosine similarity ranking via embedding_client.py with Ollama/OpenAI auto-detection; falls back to BM25 with a visible warning when no embedding server is available

find_references fix: Now uses word-boundary regex (\b) instead of substring matching, eliminating false positives (e.g., searching for io no longer matches action).

Search result caching: Results are cached with a 20s TTL via the existing FileCache L1 layer.

1.3 Reviewer (1 Tool) — Code Quality Check

Automatically check code quality before submitting a PR. Integrates with ESLint and go vet.

1.4 Tester (2 Tools) — Test Generation and Coverage

Detects function signatures to automatically generate test templates and measure coverage.

1.5 Deep Analyzer (4 Tools) — Deep AST Analysis

Tools: analyze_call_graph, map_dependencies, extract_patterns, reverse_engineer

Analyzes call graphs, dependencies, and recurring patterns using tree-sitter AST, and automatically generates documentation.

1.6 Whiteboard (4 Tools) — AI-Human Visual Collaboration

Tools: draw_on_whiteboard, get_whiteboard_state, capture_screen, check_uploaded_files

The AI draws on a Fabric.js canvas, reads user modifications, captures the screen, and monitors files uploaded to the Dropzone.

1.7 Fix Loop (3 Tools) — Autonomous Build & Fix Loop

If a build fails, the LLM automatically analyzes the error, looks up past fix patterns in Crow Memory, and suggests fixes. Supports Human-in-the-Loop.

New fix_loop intent — the ux_coordinator can now route bug-fix and error-recovery scenarios directly to the Fix Loop workflow, enabling seamless transitions from intent detection to autonomous repair.

1.8 Integrated (1 Tool) — Unified Scenario Tools

Combines multiple tools into a single workflow. Just say "Review this", and it runs search → review → quality → patterns sequentially.

1.9 Analysis (4 Tools) — Code Explanation and Diff Analysis

Explains what specific code lines do, analyzes git diffs, supports PR reviews, and proposes bulk refactoring.

1.10 Knowledge & Preferences (3 Tools) — Project Knowledge Memory

recall_project recalls auto-captured project knowledge (learned at bridge startup via _auto_learn_project), while learn_preference and get_preferences store and retrieve your coding style and preferences in Crow Memory.

1.11 Web (2 Tools) — Web Search and Page Analysis

Used to reference external documentation or search for the latest technical information.

v0.16.0 improvements:

• web_search now falls back to DuckDuckGo when EXA_API_KEY is absent, ensuring search always works out of the box
• Errors are surfaced with structured error codes instead of silent except: return []
• Retry logic: 2 retries with exponential backoff (0.5s, 1.5s)

1.12 SSA (1 Tool) — Spatial Statistical Analysis

Spatial Statistical Aggregator: OpenCV-based image pixel statistics analysis, including OCR.

1.13 Editor (1 Tool) — AI-Safe File Editing

Apply patches to files without worrying about missing parameters. The apply_patch tool:

• path optional: Auto-detects target file from diff content
• Fuzzy matching: Auto-corrects up to 85% similarity (ignores whitespace/indentation differences)
• AST-Guided Smart Ellipsis: Detects // ..., # ..., /* ... */ placeholders in SEARCH blocks and resolves them via text fuzzy matching + optional AST scope verification — LLMs can safely use // ... existing code ... placeholders
• Transactional Apply: All SEARCH/REPLACE blocks are validated in an in-memory dry-run first (Phase 1). Only if all blocks succeed does the atomic commit (Phase 2) proceed; on any failure, the entire operation rolls back — no partial modifications
• Auto backup: Backs up to ~/.vibezoo-backup/ before modification
• Supports both ======= / -------: Compatible with apply_diff

1.14 Setup (1 Tool) — Automation

Installs VibeZoo dependencies and auto-configures MCP/Zoo settings. ripgrep auto-install: When running vibezoo_setup(target="full"), automatically installs ripgrep via winget/choco/scoop or direct download.

---

2. Vision AI Pipeline

VibeZoo includes a built-in Vision AI pipeline for image analysis:

• MiniCPM-V: Lightweight Vision-Language Model. Performs image captioning and analysis locally (GGUF + llama-cpp-python).
• OCR: Tesseract / PaddleOCR for text extraction (supports Korean, English, Japanese, Chinese).
• SSA: OpenCV-based Spatial Statistical Aggregator for pixel statistics and spatial pattern analysis.

---

3. Crow Memory Overview

You can find the standalone Crow Memory repository and core engine here: vibezoo/crowmemory

3.1 Philosophy — "Crow remembers not the code, but the hand that wrote it."

Transformer-based LLMs are fixed at training time. RAG or SQLite-based solutions stack information infinitely like a "100% accurate notepad," but the human brain doesn't work that way. Forgetting is not a bug. Crow's fixed-size weight matrices and λ (decay rate) implement "Creative Forgetting." By giving up 100% accurate recall, the AI responding through Crow becomes biased toward the current you.

3.2 The 8 Registers

• Code Domain: style, bug, arch, context
• Life Domain: life_pref, life_avoid, life_phil, life_context

3.3 Hebbian EMA Update Rule

Crow updates weights using an Exponential Moving Average (EMA) based on Hebbian learning: W_new = λ · W_old + (1 - λ) · (key ⊗ value)

All memories are compressed into a fixed-size crow.bin (140MB). When capacity is exceeded, old memories naturally fade.

3.4 10 Crow MCP Tools

Includes tools like crow_recall, crow_ingest, crow_diagnostics, crow_manage_backup, and more to interact with the synaptic memory.

---

4. Windows Auto-Start & Stability

VibeZoo Bridge auto-starts on Windows boot, with the Watchdog continuously monitoring server health.

Auto-Start (Windows Startup)

start_vibezoo_servers.bat auto-starts the following on Windows boot:

1. Crow Memory Server (port 9020) — Built-in lock cleanup and health check
2. VibeZoo Bridge (port 9027) — Runs via absolute Python path with stderr/stdout logging
3. Health Check — Waits up to 60 seconds for each server to become ready (backoff: 2→2→3→5s)
4. Watchdog — Monitors Crow Memory and VibeZoo Bridge at 30-second intervals

VS Code Extension Auto-Connect

When the VibeZoo extension activates inside VS Code it performs the following auto-connect sequence:

1. Resolves a working Python interpreter via PythonResolver.
2. Spawns the VibeZoo MCP Bridge from the bundled extension/mcp-servers/ directory.
3. Spawns a Crow Memory fallback server (or proxies to an existing external Crow server).
4. Writes/updates global mcp_settings.json with the current Streamable HTTP endpoint and removes legacy .roo/mcp.json config.
5. Runs SelfCheck diagnostics in the background and auto-recovers Bridge/MCP failures.

Watchdog

• Crow Memory Watchdog: watch_crow_sse.bat
• VibeZoo Bridge Watchdog: watch_vibezoo_bridge.bat
  • Runs netstat + curl health check every 30 seconds
  • Auto-restarts the server if it dies (stops after 5 consecutive failures)
  • Logs to: %USERPROFILE%\.vibezoo\watchdog_bridge.log

REST API (Crow Memory Integration)

VibeZoo Bridge stores and retrieves memories via Crow Memory's REST API:

• GET /health — Check server status
• POST /ingest — Store errors and context
• GET /recall — Search for similar errors and patterns

Changelog

• v0.16.0 (2026-09-02):
  • Tool inventory cleanup (39→33): removed aggregate tools find_bugs, suggest_refactor, generate_docs, learn_project, auto_analyze_whiteboard, auto_analyze_after_drop — equivalent workflows are now prompt-level compositions of the remaining tools
  • Deleted dead github_diver.py; purged ghost tool read_project_file
  • analyze_uploaded_file now supports dropzone session tracking (track_dropzone parameter)
  • Removed extension wrapper commands vibezoo.findBugs / vibezoo.suggestRefactor / vibezoo.generateDocs / vibezoo.learnProject and orphaned NLS keys across 20 locales
  • Path-safety fixes: no absolute/user-specific paths in bridge output and docs
• v0.15.2 (2026-07-25):
  • Tool Ecosystem Overhaul: Comprehensive enhancement of the VibeZoo MCP tool ecosystem
  • New modules: fuzzy_matcher.py (trigram fuzzy matching), embedding_client.py (embedding-based semantic search), ast_singleton.py (shared AST singleton)
  • search_codebase(mode="fuzzy") now performs real trigram approximate matching (was identical to auto)
  • search_codebase(mode="semantic") now uses embedding-based cosine similarity ranking with BM25 fallback
  • find_references fixed: word-boundary regex eliminates false positives
  • Search result caching with 20s TTL
  • web_search DuckDuckGo fallback when EXA_API_KEY absent
  • Dead code cleanup: 12 dead entries removed from _tool_registry (20→8)
  • Tool consolidation: 5 duplicated AST singletons → shared ast_singleton.py
  • max_tokens truncation now works in 5 tools
  • 104 CI tests pass
• v0.15.1 (2026-06-16):
  • Standard path migration: runtime directory changed to %USERPROFILE%\mcp-servers\vibezoo\ (Windows) / ~/mcp-servers/vibezoo/ (macOS/Linux)
  • init_vibezoo.bat / init_vibezoo.sh now copy bridge files to the standard target directory
  • autoStartCommand updated to cd /d "%USERPROFILE%\mcp-servers\vibezoo" && start_vibezoo_bridge.bat
  • Fixed auto-start process conflict: preserved autoStart and autoStartCommand configuration for Zoo Code auto-start capability. Added physical port occupancy check (netstat/lsof) and aggressive zombie cleanup to prevent winerror 10048 socket bind errors.
• v0.15.0 (2026-06-13):
  • Auto-connect fundamental fix: always write .roo/mcp.json even when global MCP config has a vibezoo entry
  • New PythonResolver, McpConfigService, VscodePaths
  • Bundled Python bridge and Crow fallback server inside the VSIX (extension/mcp-servers/)
  • Replaced crow_memory_server.py stub with a real HTTP fallback/proxy server
  • SelfCheck auto-recovery for Bridge and MCP configuration failures
• v0.00.1 (2026-06-12):
  • MCP server stabilization: REST API integration, full startup batch rewrite, Watchdog introduction
  • Broad except Exception → specific exception handling
  • Retired FAKE Crow server → unified on real Crow
  • Removed hardcoded Crow URLs → uses config.CROW_URL

---

5. Quick Start

5.1 Requirements

• Python 3.10+
• Zoo Code (or MCP-compatible AI Coding Agent)
• Git

5.2 Installation

See the Out-of-the-Box Setup section above, or docs/INSTALLATION.md for the full guide:

git clone https://github.com/vibezoo/VibeZoo_forZoocode.git
cd VibeZoo_forZoocode
# Windows: init_vibezoo.bat   |   macOS/Linux: bash init_vibezoo.sh

5.3 Manual VibeZoo MCP Bridge Configuration

// global mcp_settings.json
{
  "mcpServers": {
    "vibezoo-bridge": {
      "url": "http://localhost:9027/mcp",
      "type": "streamable-http"
    }
  }
}

---

6. Architecture Overview

┌──────────────────────────────────────────────────────────────┐
│                      VS Code Window                           │
│                                                               │
│  ┌──────────────────────┐    ┌────────────────────────────┐  │
│  │   Zoo Code (LLM)     │    │  VibeZoo Extension          │  │
│  │                      │    │                             │  │
│  │  • LLM Reasoning     │    │  • FixLoopManager          │  │
│  │  • Built-in Crow     │    │  • VisualVibePanels        │  │
│  │    (localhost:9020)  │    │  • Safety Net (Guard.git)  │  │
│  │  • MCP Client        │    │                            │  │
│  └──────────┬───────────┘    └─────────────┬──────────────┘  │
│             │ MCP/SSE                      │ child_process    │
└─────────────┼──────────────────────────────┼──────────────────┘
              ▼                              ▼
┌─────────────────────────┐  ┌─────────────────────────────────┐
│ Crow Memory (9020)      │  │ VibeZoo MCP Bridge (9027)       │
└─────────────────────────┘  └─────────────────────────────────┘

Guard.git: VibeZoo's Guard.git feature prevents AI agents from accidentally running rm -rf * / rmdir /s /q etc. and deleting the .git folder. It provides OS-level ACL protection using Windows icacls, Linux chattr, macOS chmod, multi-root workspace support, Git Worktree compatibility, FileSystemWatcher real-time monitoring, Yocto snapshot backup, and SelfCheck integrity diagnostics.

---

7. License and Contact

MIT License — See LICENSE.

Commercial Services / Custom Development

Crow Memory is free to use under the MIT license. However, organizations have different needs—security, proprietary LLM integration, custom encoding, compliance, etc.

We provide custom development services:

• 🔒 Secure Deployments (Air-gapped, encrypted crow.bin, RBAC)
• 🏢 Enterprise Customization (Custom registers, industry-specific decay profiles)
• 🤖 LLM-Specific Optimization (Fine-tuned embeddings, optimized weight matrices)
• 🧩 Software Integration (Non-VS Code IDEs, CI/CD hooks)
• 📊 Enterprise Analytics (Memory usage dashboards, drift alerts)

Contact us for enterprise, secure, or custom-tailored Crow Memory development: 📧 myk1yt@gmail.com

---

VibeZoo v0.16.0 — July 2026 Co-designed by Stefano, Kim & AI