JAGGU
Developer-Controlled AI Coding Agent for VS Code
Plan precisely. Stage safely. Verify with tests. Never surrender control.
TypeScript |
React |
VS Code API |
Hugging Face |
Ollama
⚡ Overview
Most AI coding assistants either output code snippets to a chat window for you to manually copy-paste, or silently modify your workspace files with opaque, unpredictable changes.
JAGGU delivers true autonomous engineering without sacrificing developer control.
Built directly on the VS Code extension runtime, JAGGU indexes your codebase with Ripgrep, synthesizes structured multi-phase execution plans, stages surgical diffs in an in-memory shadow buffer, runs your project's test suites to verify its solutions, and autonomously repairs compiler or test errors before presenting the finished work for your final approval.
Zero Middleman Proxy • 100% Direct TLS Connections • Air-Gapped Local Model Support
🏛️ Architecture Diagram
JAGGU uses a decoupled, event-driven multi-layer architecture:
┌─────────────────────────────────────────────────────────────────────────┐
│ Visual Studio Code │
│ (Sidebar Webview React 18 • Native Diff Editor • Status Bar) │
└────────────────────────────────────┬────────────────────────────────────┘
│ Extension Host Typed IPC
┌────────────────────────────────────▼────────────────────────────────────┐
│ JAGGU Extension Host │
│ (SidebarProvider • SecretStorage • VirtualDocProvider) │
└────────────────────────────────────┬────────────────────────────────────┘
│
┌────────────────────────────────────▼────────────────────────────────────┐
│ Agent Orchestrator │
│ (Deterministic 7-State Finite State Machine / Safety Model) │
├────────────────────────────────────┬────────────────────────────────────┤
│ Context Engine │ Verification Engine │
│ (Ripgrep Index + AST RepoMap) │ (Test Runners + LSP Diagnostics) │
├────────────────────────────────────┴────────────────────────────────────┤
│ Tool Executor │
│ (ReadFile • ProposeEdit • ApplyEdit • ExecuteTerminal • Git) │
└────────────────────────────────────┬────────────────────────────────────┘
│ Direct TLS / Local IPC
┌────────────────────────────────────▼────────────────────────────────────┐
│ Polymorphic Model Gateway │
│ ┌─────────────────────┬──────────────────────┬────────────────────┐ │
│ │ Cloud Models │ Air-Gapped Ollama │ OpenAI-Compatible │ │
│ │ (Claude/GPT/Gemini) │ (Qwen/DeepSeek-R1) │ (vLLM / LM Studio) │ │
│ └─────────────────────┴──────────────────────┴────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
⚡ Install Extension
Quick Install via VS Code Terminal
Run the following command in your terminal to install the packaged extension directly into your local VS Code instance:
code --install-extension jaggu-vscode-0.1.0.vsix
Install via VS Code UI
- Open VS Code.
- Open the Command Palette (
Cmd+Shift+P on macOS / Ctrl+Shift+P on Windows/Linux).
- Type
Extensions: Install from VSIX... and press Enter.
- Select
jaggu-vscode-0.1.0.vsix.
✨ Key Features
| Capability |
What It Does |
Why It Matters |
| 🛡️ Controlled Autonomy |
Deterministic 7-state Finite State Machine (IDLE ➔ PLANNING ➔ APPROVAL ➔ EXECUTING ➔ VERIFYING ➔ SELF-HEAL ➔ COMPLETE) |
No runaway loops or silent background file corruption. Every transition is transparent and observable. |
| 📋 Interactive PlanCards |
Generates structured, milestone-based execution plans |
Review and approve the agent's strategy before a single line of code is staged. |
| 🪟 Shadow Buffer Diffs |
Edits staged in an in-memory VirtualDocStore |
Zero direct file overwrites. Inspect changes side-by-side using VS Code's native diff editor. |
| 📁 Selective File Approval |
Granular per-file Accept / Reject gates |
Approve only the files you want, reject unwanted changes, or iterate on the diff before committing. |
| 🔄 Automated Test Verification |
Auto-detects & runs test runners (Jest, Vitest, PyTest, Go, Cargo) |
The agent never marks a task finished without empirical test proof. |
| 🩹 Self-Healing Loop |
Catches exit codes, stderr & LSP compiler diagnostics |
Automatically attempts up to 3 targeted repairs if tests fail or syntax errors arise. |
| 🔒 Git Safety Checkpoints |
Non-destructive working tree safety stashes |
Protects unstaged work and provides instant rollback points. Never pushes to remote. |
| 🌐 Provider Agnostic & Local AI |
Native support for Claude 3.5/3.7, GPT-4o, Gemini 2.0, Hugging Face, Ollama, & vLLM |
Complete independence from vendor lock-in. Run 100% offline and air-gapped with zero telemetry. |
🎮 Using JAGGU
- Open the Sidebar: Click the JAGGU coding glasses icon in the Activity Bar (or run
JAGGU: Open Chat).
- Configure Your Provider:
- Local AI (Ollama): Ensure Ollama is running (
ollama serve). Select ollama as your provider and type your model (e.g. qwen2.5-coder:7b). No API key needed!
- Cloud Models: Run
JAGGU: Manage API Keys (SecretStorage) from the Command Palette, pick your provider, and paste your API key.
- Submit a Task: Type your request (e.g. "Fix the clock skew error in the JWT auth middleware and add regression tests").
- Approve the Plan: Review the milestone PlanCard and click Approve.
- Inspect Diffs & Review: Check the staged changes in VS Code's side-by-side diff viewer and accept or reject files individually.
⌨️ VS Code Commands
Access all JAGGU capabilities directly via the Command Palette (Cmd+Shift+P / Ctrl+Shift+P):
| Command |
Identifier |
Action |
JAGGU: Open Chat |
jaggu.openChat |
Focuses and opens the JAGGU sidebar view |
JAGGU: Start New Session |
jaggu.startSession |
Resets conversation state and initializes a clean task session |
JAGGU: Cancel Active Task |
jaggu.cancelSession |
Instantly aborts in-flight generation, streaming, or test execution |
JAGGU: Set Hugging Face Token |
jaggu.setHuggingFaceToken |
Stores your free Hugging Face token in SecretStorage |
JAGGU: Remove Hugging Face Token |
jaggu.removeHuggingFaceToken |
Clears Hugging Face token from SecretStorage |
JAGGU: Test Hugging Face Connection |
jaggu.testHuggingFaceConnection |
Tests live connectivity to Hugging Face inference router |
JAGGU: Check Provider Status |
jaggu.checkProviderStatus |
Verifies runtime health of local and cloud providers |
JAGGU: Manage API Keys (SecretStorage) |
jaggu.setApiKey |
Securely stores API keys in OS Keychain / Credential Manager |
JAGGU: Select Model Provider |
jaggu.selectProvider |
Quickly switches active AI provider (Hugging Face, Ollama, Anthropic, etc.) |
JAGGU: Select Active Model |
jaggu.selectModel |
Selects specific model tag or enters custom model string |
JAGGU: Approve Plan |
jaggu.approvePlan |
Approves staged milestone engineering plan |
JAGGU: Reject Plan |
jaggu.rejectPlan |
Rejects proposed plan and allows prompt refinement |
JAGGU: Approve Changes |
jaggu.approveEditSet |
Atomically applies reviewed change set to workspace |
JAGGU: Reject Changes |
jaggu.rejectEditSet |
Rejects change set with zero disk mutations |
JAGGU: Review Proposed Diff |
jaggu.reviewDiff |
Opens native VS Code side-by-side diff review |
🤖 Supported Models & Providers
JAGGU connects directly to AI providers with zero intermediate proxy servers and Free-First Auto Routing:
| Provider |
Access Tier |
Highlighted Models |
Privacy / Connection |
| Auto (Default) |
Free / Local First |
Dynamic best model (Qwen3-Coder-30B, qwen2.5-coder:7b) |
Automated Zero-Cost Priority |
| Hugging Face |
Free Cloud |
Qwen/Qwen3-Coder-30B-A3B-Instruct, meta-llama/Llama-3.1-8B |
Direct Client-to-API TLS |
| Local Ollama |
100% Local / Free |
qwen2.5-coder:7b, deepseek-r1:8b, llama3.3:70b |
100% Offline / Air-Gapped |
| OpenAI-Compatible |
Local Server |
Any self-hosted model (vLLM, LM Studio, LocalAI) |
Local / Private Network |
| Anthropic |
Paid (Opt-in) |
claude-3-7-sonnet-latest, claude-3-5-sonnet-latest |
Direct Client-to-API TLS |
| OpenAI |
Paid (Opt-in) |
gpt-4o, gpt-4o-mini, o1, o3-mini |
Direct Client-to-API TLS |
| Google Gemini |
Paid (Opt-in) |
gemini-2.0-flash, gemini-1.5-pro |
Direct Client-to-API TLS |
| Mock Engine |
Deterministic |
mock-fast, mock-reasoning |
In-memory simulated for testing |
🛡️ Security & Privacy First
- 🔐 OS-Level Secret Storage: API keys are saved exclusively into VS Code's
SecretStorage API (backed by macOS Keychain, Windows Credential Manager, or Linux Secret Service). Never saved to plain text or settings files.
- 🧹 Secret Sanitizer: Built-in regex sanitization strips API keys, tokens, and sensitive headers from debug logs and error messages.
- 🛡️ Zero Telemetry Exfiltration: Code, prompts, and tokens are never sent to third-party analytics or intermediary relays.
- 🛑 Shadow Staging Protection: The agent cannot execute destructive writes directly to your disk; all diffs reside in memory until explicitly accepted.
- 🔒 Workspace Confinement: Path resolution strictly prevents directory traversal (
/../) attacks beyond the active project folder.
📄 License
This extension is licensed under the MIT License — see the LICENSE file for details. Built upon the Visual Studio Code extension architecture.
