A local autonomous coding agent CLI powered by Ollama.
Project description
🧠 DevAgent
Execution-Grounded Orchestration for Autonomous Local Coding.
🌐 Live Site • 📦 PyPI • 📜 Docs
DevAgent is a research-grade, local-first coding agent runtime designed to bridge the gap between LLM-generated logic and production-ready execution integrity.
Quick Start • Architecture • Safety & Containment • Benchmarks • Troubleshooting
🛡️ The Problem: The "LLM-Execution Gap"
Most autonomous coding agents fail because they operate in a vacuum. They generate code that looks correct but fails at runtime due to environment drift, dependency conflicts, or invalid execution assumptions.
DevAgent is Execution-Grounded Orchestration. It doesn't just guess code; it manages the entire lifecycle of a fix:
- Discovery: Scans and maps the environment.
- Isolation: Provisions a clean, sandboxed virtual environment.
- Repair: Autonomously resolves missing dependencies.
- Validation: Verifies patches against your actual test suite.
🏗️ High-Integrity Architecture
DevAgent v3.4.1 implements a multi-layer orchestration stack designed for reliability over hype.
graph TD
subgraph "Orchestration Layer"
CLI[DevAgent CLI] --> Planner[Task Planner]
Planner --> Retrieval[Hierarchical Retrieval: FAISS + Ripgrep]
Retrieval --> Agent[ReAct Agent: Thought/Action/Observation]
end
subgraph "Execution Layer"
Agent --> Patch[Surgical Patch Engine: Line-Level Diffs]
Patch --> Reviewer[Self-Review Loop: APPROVED/REVISE]
Reviewer --> Runtime[Environment Runtime]
end
subgraph "The Maturity Layer"
Runtime --> Discovery[Auto-Dependency Discovery]
Discovery --> Isolation[Venv Isolation: .tmp_envs]
Isolation --> Repair[Autonomous Repair Loop: repair_environment]
Repair --> Validation[Validation Engine: pytest]
end
Validation -- Pass --> Apply[Apply to Root Repo]
Validation -- Fail --> Agent
🚀 Quick Start
1. Installation
Install the CLI via PyPI. Ensure Ollama is running locally.
pip install devagent-cli
2. Verify Infrastructure
Check your local environment, connectivity, and dependency health.
devagent doctor
3. Run Your First Task
Execute an autonomous fix on any repository.
devagent run --task "Implement input validation for the user login" --root ./my-project
✨ Advanced Features
🔍 Hierarchical Retrieval
Instead of dumping your entire codebase into a context window, DevAgent uses a multi-tier search:
- Global Map: Scans the file structure to identify relevant modules.
- Semantic Tier: FAISS-powered vector search for conceptual matching.
- Precision Tier: Ripgrep for exact symbol/error discovery.
🏖️ Environment Isolation & Repair
DevAgent is the first local agent to treat the environment as a first-class citizen. It detects requirements.txt or pyproject.toml, creates an isolated .tmp_envs/ runtime, and autonomously installs missing packages if it encounters a ModuleNotFoundError.
🩹 Surgical Patch Engine
Most agents ruin git history by rewriting entire files. DevAgent generates line-level unified diffs, applying only the necessary changes while preserving your code style, comments, and structure.
🔐 Safety-First Containment
We built DevAgent for engineers who care about their host systems.
- Dry-Run Mode: Visualize every change before it happens.
- Atomic Snapshots: A safety restore point is created before every execution.
- Instant Rollback: Revert any agent intervention with
devagent rollback. - Sandbox Isolation: Every run is contained in a separate workspace until validation passes 100%.
📊 Empirical Validation
We don't fake our success rates. DevAgent is evaluated against a public, messy benchmark suite.
| Metric | Result | Infrastructure Status |
|---|---|---|
| Dependency Repair | 95% | ✅ Production Ready |
| Unit Bugfixes | 80% | ✅ Highly Reliable |
| Refactoring | 20% | 📈 Improving (Model Bounded) |
| Isolation Safety | 100% | ✅ Absolute Containment |
Full Report: v3.4.1 Benchmark Analysis
🤝 Contributing
Built with a focus on Systems Thinking. PRs that improve orchestration reliability, environment detection, or patch precision are highly encouraged.
# Clone and Install in Editable Mode
git clone https://github.com/VedantJadhav701/Developer-Code-Intelligence-Agent.git
pip install -e .
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file devagent_cli-3.7.5.tar.gz.
File metadata
- Download URL: devagent_cli-3.7.5.tar.gz
- Upload date:
- Size: 41.2 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
bd893a2b80ea53d69f780a859679adccf9faa31d3a6bdffdcdbf7f318ca7a7b8
|
|
| MD5 |
632c1c3fb888ebef46fdd97ad28e95b9
|
|
| BLAKE2b-256 |
3a434c583af65593319ce9cd63505669971529732944cb40e1785ce91a39a053
|
File details
Details for the file devagent_cli-3.7.5-py3-none-any.whl.
File metadata
- Download URL: devagent_cli-3.7.5-py3-none-any.whl
- Upload date:
- Size: 49.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
2761ce2f4850f7e558cf75106fef6616fa6ea3f84178397e0240013fd42c3a34
|
|
| MD5 |
add20b49e56ec22af7c0b1c7eb0a57b5
|
|
| BLAKE2b-256 |
5585819d645077978fe63b7222830eb37a2dc3dc1d1705ce7d95b163f402cacc
|