📖 Overview

EdgeMind is a terminal-first autonomous software engineering assistant designed to operate entirely on local hardware.

Unlike conventional AI coding assistants that rely heavily on cloud infrastructure, EdgeMind focuses on resource-aware execution, dynamic model routing, streaming agent activity, and context-aware intent routing to perform complex software engineering tasks locally.

EdgeMind behaves like a senior software engineering companion capable of:

• 🧠 Context-Aware Intent Routing: Automatically distinguishes between code execution tasks, follow-up questions ("What did you change?", "Why?"), and conversational architecture discussions.
• ⚡ Real-Time Streaming Agent Activity: Live progress updates (● Understanding request..., ✓ Found bad.java, → Analyze → Fix → Convert to Python, etc.).
• 🤖 Intelligent Model Manager: Automatically discovers installed local Ollama models (e.g. qwen2.5-coder:7b, codellama, phi3) and uses available local models without forcing unnecessary multi-GB downloads.
• 🛠️ Autonomous Code Creation & Modification: Intelligently infers whether a new file is required (e.g. bad.java -> bad.py) versus modifying an existing file (fix bad.py).
• 🛡️ Hardened Multi-Layer Verification: Disk-level post-write inspection, syntax validation, backup exclusion, and project-root security boundary enforcement.
• 🗂️ Enriched SQLite Project Memory: Remembers execution requests, plans, diffs, analysis findings, and validation results across interactive sessions.

🎯 Vision

EdgeMind explores a central research question:

Can autonomous software engineering agents operate efficiently on local consumer hardware while intelligently adapting to limited computational resources and user intent?

The project combines:

• Edge AI & Local Inference (Ollama)
• Agentic AI (LangGraph execution graphs)
• Resource-Aware Computing
• Real-Time Streaming Agent Activity
• Persistent Project Memory (SQLite)

✨ Core Features in V2.1

1. Context-Aware Conversation & Follow-Up Routing

EdgeMind distinguishes execution requests from follow-up questions and conversational inquiries:

• Follow-Up Questions ("What did you change?", "Why did you change it?", "Explain that"): Loads previous execution context and SQLite history to explain edits without altering files or running new plans.
• Conversational Queries ("What do you think about this architecture?", "Do you think this approach is good?"): Responds conversationally using active project context as a pair-programming companion without triggering file edits.

2. Real-Time Streaming Agent Activity

Exposes safe, user-facing agent activity events live during execution:

  ● Understanding request...
  ● Identifying source file...
  ✓ Found bad.java
  ● Determining requested operation...
  → Analyze → Fix → Convert to Python
  ● Creating execution plan...
  ✓ 3 tasks planned
  ● Analyzing bad.java...
  ✓ Analysis complete
  ● Generating Python implementation...
  ✓ Generated bad.py
  ● Validating generated code...
  ✓ Python syntax valid
  ● Reviewing changes...
  ✓ Source preserved

3. Intelligent Model Manager & First-Run Setup

• Automatically detects installed local Ollama models.
• Uses existing local models (e.g. qwen2.5-coder:7b, codellama, deepseek-coder, phi3:mini) without requiring downloads.
• If no model is present, recommends a resource-aware fallback (qwen2.5-coder:3b) and prompts before downloading (Download model? [Y/n]).

4. Enriched Execution Summary & File Review

Reports comprehensive file status and validation results:

Files Status:
  Created  : bad_algorithm.py (NEW FILE)
  Modified : None
  Preserved: bad.java (UNTOUCHED)

Validation & Review:
  ✓ Source file preserved: /path/to/bad.java
  ✓ Target file created: /path/to/bad_algorithm.py
  ✓ Syntax validation passed: Validation Passed

🏗️ Architecture

                       User Prompt
                            │
                            ▼
                  Intent Router (V2.1)
           ┌────────────────┼────────────────┐
           ▼                ▼                ▼
     Conversational     Follow-Up         Execution
        Handler          Handler           Request
           │                │                │
           │           (Load Memory          ▼
           │           & SessionState)  LangGraph Engine
           │                │           (Planner -> Discover ->
           │                │            Router -> Executor ->
           │                │            Reviewer -> Memory)
           └────────────────┼────────────────┘
                            │
                            ▼
                 Activity Streamer & CLI

⚙️ Installation & Setup

Install via PyPI

pip install edgemind

To upgrade an existing installation:

pip install -U edgemind

Launch the interactive CLI shell:

edgemind

Install from Source

git clone https://github.com/Akhilesh-Venkiteswaran/EdgeMind.git
cd EdgeMind
python3 -m venv venv
source venv/bin/activate
pip install -e .

Prerequisites

1. Ollama: Download from https://ollama.com
2. Local Models: EdgeMind automatically detects any installed model. Recommended models:

   ollama pull qwen2.5-coder:3b
   

🧪 Running Tests

Run the complete test suite:

# Deterministic Unit Tests (No Ollama required)
pytest -m "not ollama" -v

# Live Ollama Integration Tests
pytest -m ollama -v

# Complete Test Suite (All 40+ tests)
pytest -v

📜 License

MIT License. Free to use, modify, and distribute.