Report Abuse

Basic Information

Devon is an open-source pair programmer designed to help developers interactively work on codebases. It provides a user-facing agent accessible via a web/Electron UI and a terminal UI that runs in the project directory so the agent can inspect and modify project files. The project requires node.js, npm and pipx for installation and accepts a single API key for model access (Anthropic or OpenAI are supported). Devon aims to assist with multi-file edits, code exploration, writing configuration and tests, fixing bugs, and exploring architecture. It supports multiple models including Claude 3.5 Sonnet, GPT4-o, Groq llama3-70b and local Ollama models, with additional model integrations planned. The repository is community-driven, licensed under AGPL, and provides telemetry that can be disabled by an environment variable.

Links

Categorization

App Details

Features
Devon includes both a graphical UI (Electron/web) and a terminal UI (devon-tui) with simple run commands such as npx devon-ui and devon-tui. Core features documented in the README are multi-file editing, codebase exploration, configuration writing, test generation, bug fixing, architecture exploration and local model support. Installation uses pipx for the backend package devon_agent and npm for UI tooling. It lists multi-model support and shows how to configure local models via ollama for a local deepseek-coder image. The README also documents limitations, mentions minimal functionality for non-Python languages, and notes local mode currently degrades performance. The project plans a plugin system, improved Electron app features and better code indexing and memory modules.
Use Cases
Devon helps developers by automating and accelerating common engineering tasks inside the context of a project. It can perform multi-file refactors, search and reference code across a repository, generate or improve tests, fix bugs identified in the codebase and assist with configuration and architecture questions. The terminal and UI workflows make it straightforward to run the agent locally while restricting access to the working directory. Multi-model and local model options can reduce latency and cost for some users. The project is intended to be extensible via a forthcoming plugin system and to improve developer productivity through better context gathering, code indexing and interactive steering features. Community support is available via Discord and contributions are welcome.

Please fill the required fields*