The Cognitive Reasoning Engine for Autonomous Development

English | 繁體中文 | Documentation

Boring-Gemini isn't just a collection of tools; it's the thinking layer for your AI development workflow. While standard AI models suggest code, Boring reasons, verifies, and learns.

"Intent is the new Implementation."

In the era of Vibe Coding, your role shifts from writing syntax to defining Intent. Boring-Gemini acts as your agentic partner, handling the gap between a "Vibe" (Natural Language) and "Production" (Verified Code).

Boring implements a rigorous Planning -> Execution -> Verification loop. It doesn't just run commands; it uses sequentialthinking and criticalthinking to analyze its own steps, critiquing logic before hitting your disk.

The first agent with a Global Brain. When Boring encounters a failure, it consults its persistent knowledge base (.boring/brain) to recall how similar issues were solved across sessions. It learns from its mistakes so you don't have to.

Designed for the modern Python stack. Boring natively supports uv for near-instant package management, lockfile synchronization, and isolated environment execution.

Trust is built on safety. Boring automatically creates Git Checkpoints before any risky operation. Combined with Shadow Mode, you have a "undo" button for AI agentic actions, ensuring your repository remains stable even during complex refactors.

Boring achieved Full-Power status by activating all high-value cognitive tools.

• SpecKit Activation: Enabled full Specification-Driven Development tools (plan, tasks, analyze) for methodical planning.
• Global Brain Tools: Unlocked cross-project knowledge sharing (boring_global_export) to recycle success patterns.
• Skills Autonomy: New boring_skills_install allows the Agent to autonomously install missing Python packages.
• Node.js Autonomy: Automatic Node.js download/install to ensure gemini-cli works even on fresh systems.
• Lightweight Mode (BORING_LAZY_MODE): Perfect for "Quick Fixes" without polluting directories with .boring folders.

Introduction of Self-Awareness and Adaptive Safety.

• Usage Dashboard (P4): The Agent now tracks its own tool usage, visualizing stats in a CLI/Web dashboard.
• Anomaly Safety Net (P5): Automatically halts "stuck" loops (same tool + same args > 50 times) to save tokens and prevent crashes.
• Contextual Prompts (P6): Adaptive Profile now injects specific guides (e.g., Testing Guide) only when you need them.

The legacy production standard. Introduction of a State-Machine Workflow and Deep Shadow Mode.

• Unified Flow Graph: Dynamic orchestration (Architect -> Builder -> Healer -> Polish -> Evolver).
• Deep Shadow Mode: HealerNode activates STRICT safety enforcement during auto-repairs.
• Cognitive Reflex: Semantic Search integrated into the Brain for fuzzy error correction.

The current high-performance standard. Focused on Parallel Orchestration and Hybrid Storage.

• Async Agent Runner: Parallel execution of sub-tasks for faster completion.
• Semantic Storage Fallback: FAISS integration for robust vector search when ChromaDB is unavailable.
• One Dragon 2.0: Enhanced state management for complex "One Dragon" workflows.

The cutting edge. Shifting from reactive to Predictive and Local-First.

• Offline-First Mode: Local LLM support (llama-cpp-python) for 100% privacy and zero-network operation.
• Predictive Error Detection: AI-powered anti-pattern detection and proactive warnings before errors occur.
• AI Git Bisect: Semantic analysis of commit history to instantly identify bug sources.
• Lazy Loading System: Optimized MCP startup (<500ms) for high-performance developer environments.

The Sovereign Edition. Achieving "100-Point Perfection" through enterprise-grade governance and knowledge continuity.

• Project Biography (boring bio): Persistent team-wide knowledge continuity.
• Policy-as-Code (boring policy): Secure, auditable tool permission guardrails.
• Sovereign Audit (boring perfection): Automated 7-stage validation for production readiness.
• Self-Healing Doctor (boring doctor --fix): Autonomous environment repair and dependency resolution.
• System Optimizer (boring doctor --optimize): Deep storage cleanup (VACUUM), brain pattern maintenance (Decay/Pruning), and reconciler checkpointing.
• Migration Engine (boring migrate): Forward-compatible project state transition.

The Delight Update. Transforming the user experience from "Frustrating" to "Fluid".

• Visual Feedback: Rich Spinners, Step Tracking (Step 5/50), and Task Progress bars.
• Cost Awareness: Proactive API cost warnings to prevent bill shock.
• Resilience: Robust file locking handling (WinError 32) and friendly error translations.
• State Serialization: Ability to pause and resume long-running flows.

The Democratization Update. A fully decentralized, zero-cost plugin economy.

• Pack & Install: boring pack and boring install turn any repo into a plugin.
• Knowledge Share: boring brain export lets you share your AI's learned wisdom.
• GitOps Sync: boring sync enables serverless team collaboration via Git.
• Registry Spec: Open standard (registry.json) for a decentralized internet of agents.

Boring adapts to your environment to save tokens and context:

• LITE (Default): Essential tools for daily coding (43 tools) using ~15% of context window.
• FULL: All 67+ tools active.
• ADAPTIVE (Recommended): Automatically builds a custom profile based on your top 20 most frequently used tools + Prompt Injection.
  • Enable: export BORING_MCP_PROFILE=adaptive

Boring supports multi-channel task notifications to keep you informed:

• Desktop: Windows Toast, macOS, Linux notifications.
• Webhooks: Slack, Discord.
• Messaging: LINE Notify, Facebook Messenger.
• Email: Gmail (via SMTP).

Configure these in .boring.toml:

[boring]
slack_webhook = "..."
discord_webhook = "..."
line_notify_token = "..."
gmail_user = "..."
gmail_password = "..."
email_notify = "..."

Boring is a Hybrid Agent that adapts to your workflow. It works in two distinct modes:

• Where: Inside Cursor, Claude Desktop, or VSCode.
• Role: Your "second brain". It sits alongside you as an MCP server.
• Usage: You chat with it via the @boring command. It remembers context, provides tools, and offers "Active Recall" suggestions.
• Best For: Daily coding, debugging, and interactive problem solving.

• Where: Standard Terminal / Command Prompt.
• Role: Your "unattended worker". It runs as a standalone process.
• Usage: Run boring start (NOT boring run). It reads task.md, executes the plan loop (Plan -> Code -> Test -> Fix), and stops when done.
• Best For: Bulk refactoring, massive migrations, or long-running tasks while you sleep.

• Where: extensions/vscode-boring/
• Role: A graphical interface for the CLI Agent inside VS Code.
• Features: One-click Start/Stop, Live Dashboard, and Status Bar integration.
• How to use: Open the directory in VS Code, press F5 to debug, or compile with npm run compile.

Your data stays YOURS. Boring uses a two-tier memory system:

1. Project Memory (.boring/memory/):

   • Lives in your project folder.
   • Contains project-specific graphs and indexes.
   • Safe to commit: Share with your team via Git.

2. Global Brain (~/.boring/brain/):

   • Lives in your user home directory (Local Manager).
   • Stores learned patterns and skills across all your projects.
   • NEVER uploaded: This data never leaves your machine unless you explicitly configure a sync.

Designed for Vibe Coders. Setup in < 30 seconds.

Windows (PowerShell):

powershell -c "irm https://raw.githubusercontent.com/Boring206/boring-gemini/main/scripts/install.ps1 | iex"

Linux / macOS:

curl -fsSL https://raw.githubusercontent.com/Boring206/boring-gemini/main/scripts/install.sh | bash

pip install boring-aicoding
boring wizard

# Optional: Install RAG intelligence for full semantic search
pip install sentence-transformers chromadb

uv pip install "boring-aicoding[all]"

pip install "boring-aicoding[vector]" # RAG Support
pip install "boring-aicoding[gui]"    # Dashboard
pip install "boring-aicoding[mcp]"    # MCP Server

Just say these phrases to the AI in your IDE (Cursor/Claude):

• boring_flow: 🐉 One Dragon Engine. The ultimate autonomous workflow. Handles Setup -> Plan -> Build -> Polish automatically via code.
• start session: 🚀 Vibe Session. Activates Deep Thinking to autonomously manage the entire lifecycle of a complex task.
• /vibe_start: Kick off a new project from scratch.
• quick_fix: Automatically repair all linting and formatting errors.
• review_code: Request a technical audit of your current file.
• smart_commit: Generate a semantic commit message from your progress.
• boring_vibe_check: Run a comprehensive health scan of the project.
• boring doctor --optimize: Perform deep system optimization (Storage, Brain, and Checkpoints).

Boring comes bundled with elite tools to boost AI performance:

• Context7: Real-time documentation querying for the latest libraries.
• Thinking Mode: Forces the agent into deep analytical reasoning (Sequential Thinking).
• Security Shadow Mode: A safety sandbox that intercepts dangerous AI operations.

• License: Apache 2.0
• Repository: GitHub
• Smithery: Boring Server

_{Built by Boring206 with 🤖 AI-Human Collaboration}