🌐 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
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.
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
Install the CLI via PyPI. Ensure Ollama is running locally.
pip install devagent-cliCheck your local environment, connectivity, and dependency health.
devagent doctorExecute an autonomous fix on any repository.
devagent run --task "Implement input validation for the user login" --root ./my-projectInstead 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.
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.
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.
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%.
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
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 .