hotovo/aider-desk

What it solves

AiderDesk provides a transparent and steerable orchestration layer for AI-assisted development, moving beyond a simple GUI for the Aider CLI. It solves the problem of "autopilot" AI coding assistants that lack transparency, making it easier for professional software engineers to review, control, and isolate experiments without disrupting their existing IDE and Git workflows.

How it works

AiderDesk acts as a management layer that integrates with various LLM providers (OpenAI, Anthropic, Gemini, DeepSeek, Ollama, etc.) and the Aider CLI. It uses Git worktrees to isolate tasks into separate directories, preventing the AI from breaking the active local branch. It employs a vector memory engine (LanceDB) for semantic search and intelligent repository mapping to ensure the AI only loads necessary files. The platform is extensible via TypeScript functions for custom tools, React components for UI widgets, and the Model Context Protocol (MCP) for connecting to external data sources.

Who it’s for

Professional software engineers who want high-level control over AI coding assistants, preferring a "junior pair programmer" approach over a full autopilot, and those who need to manage multiple codebases and complex tasks simultaneously.

Highlights

• Git Worktree Isolation: Each task gets its own directory for experimentation and refactoring, which is then merged back into the main flow.
• Rich Review Gates: Includes a side-by-side diff viewer for line-by-line inspection and strict tool approval gates for shell commands and file operations.
• Task & History Control: Ability to duplicate or fork tasks to explore alternative paths and surgically delete messages from chat history to clean the context window.
• Deep Extensibility: Supports lifecycle hooks, custom React UI components, and MCP client/server capabilities to integrate with enterprise APIs and internal wikis.
• Coded Memory Engine: Uses vector embeddings for semantic search and the ability to pin documentation URLs and code symbols.
• Multi-Model Orchestration: Supports 30+ LLM providers and allows the definition of specialized subagents with custom system prompts.