AiderDesk
Desktop GUI for Aider AI pair programming
Install with Homebrew
brew install --cask hotovo-aider-deskQuick Take: AiderDesk
AiderDesk is one of the most actively developed open AI coding desktops in August 2026, free shell, ferocious release velocity, and real extension depth. It rewards engineers who want steerable multi-model agents without renting an entire IDE, provided you budget API spend and learn its task/worktree model.
Best For
- •Developers who want BYO-key agentic coding on Mac
- •Teams experimenting with multi-model workflows and worktrees
- •Aider CLI users who want a full desktop control plane
What is AiderDesk? - Complete Guide for Mac Users in 2026
AiderDesk (hotovo/aider-desk) is an open desktop platform for AI-powered software engineering, born as a GUI/control plane around the Aider pairing style of agentic coding and grown into a full task/agent workbench. It targets developers who want steerable agents, multi-model providers, extensions, worktrees, and usage visibility without being locked into a single IDE subscription. August 2026 status: The project is highly active. GitHub Releases show a fast train through mid/late summer 2026, with v0.77.0 (early August 2026) as a recent highlight: optimized system prompts for frontier models, read-only server mode, live transcription via gpt-live-transcribe, streaming UX fixes, and extension API growth. Earlier summer releases added AI SDK v7 migration, tool-call streaming, ACP (Agent Client Protocol) support for IDE embedding, task scheduler/skill gallery extensions, GitHub Copilot extension paths, Cursor integration experiments, and warnings about Anthropic Agent SDK subscription policy changes (prefer standard API keys for production). Pricing model: The app itself is free to download (macOS arm64/x64 DMGs on GitHub Releases). You pay for LLM usage via BYO keys or linked subscriptions (Anthropic, OpenAI GPT-5.6-class models, Gemini, local Ollama, Copilot-related extensions, etc.). Competitors now include Cursor, Claude Code, Codex-in-ChatGPT desktop, Windsurf under Cognition, Cline/Roo-style IDE agents, and Meta's newer Muse Code agent, AiderDesk differentiates on open orchestration, extensions, and not forcing a single closed agent runtime.
close look: Why GUI Wrappers for AI Coding Tools Matter
The gap between AI coding capability and developer adoption, and how visual interfaces bridge it.
History & Background
Aider was created by Paul Gauthier in 2023 as a terminal-based AI pair programming tool. It quickly gained a reputation for best-in-class edit accuracy: it could modify multiple files in a single conversation turn, create clean git commits, and handle complex refactoring tasks. But adoption was limited to developers comfortable with terminal workflows. Many developers, especially those coming from GUI-native environments (Xcode, Android Studio, Visual Studio), found the terminal interface too unfamiliar. Hotovo built Aider Desk to bridge this gap: same engine, different interface.
How It Works
Aider Desk is built on Electron (for cross-platform GUI), wrapping Aider's Python-based CLI engine. The app communicates with Aider through its API mode, sending conversation messages and file context updates, and receiving edit operations and diff output. The visual diff rendering uses a standard diff library to convert Aider's unified diff output into a syntax-highlighted side-by-side or inline view. Project configurations are stored as JSON files locally, with conversation history indexed for search.
Ecosystem & Integrations
Aider Desk inherits Aider's broad model ecosystem. Any model that Aider supports (through direct API or Ollama) works in Aider Desk. The Aider community's prompt engineering patterns, .aider.conf.yml configuration tips, and model comparison benchmarks apply directly. Aider's plugin system for custom edit formats is also accessible through Aider Desk's settings. The app integrates with the macOS Keychain for secure API key storage.
Future Development
Aider Desk's roadmap tracks Aider's feature development. As Aider adds new capabilities: architect mode for planning before editing, linting integration, support for new models: Aider Desk exposes them through the GUI. Planned Aider Desk-specific features include inline code annotations (highlighting AI-suggested changes directly in a mini-editor), team collaboration (sharing conversation contexts between developers), and deeper git integration with branch comparison views.
Key Features
Visual File Context Management
Aider's output quality depends heavily on which files are in the AI's context window. In the CLI, you add files by typing paths: workable for 2-3 files, tedious for 15. Aider Desk shows your project's file tree with checkboxes. Click to add a file, click again to remove it. Files currently in context are highlighted with a color indicator. You can see at a glance exactly what the AI can see and modify. This visual management is the single biggest workflow improvement over the CLI for complex multi-file sessions.
Read-Only Context Files
Not every file the AI needs to see should be editable. Interface definitions, type files, configuration schemas: the AI needs to understand them but shouldn't modify them. Aider Desk lets you add files as read-only: the AI sees the content for context but the file is locked from modifications. In the CLI, you use the /read-only command. In Aider Desk, it's a toggle next to each file in the context panel.
Visual Diff Review
When the AI proposes changes, Aider Desk renders a side-by-side or inline diff with syntax highlighting. You see exactly what lines are being added, removed, or modified: the same way you'd review a pull request on GitHub. This is more readable than the terminal's inline diff output, especially for large changes spanning multiple files. You can approve or reject changes before they're applied to your working tree.
One-Click Git Operations
Aider creates a git commit for every change it makes. Aider Desk shows a visual commit history with diffs for each commit. If the AI made a change you don't like, click the revert button to undo that specific commit. No need to remember git commands or figure out which commit hash to revert. The safety net is always visible and one click away.
Multi-Model Switching
Configure multiple LLM providers in settings and switch between them per conversation. Start a session with Claude for a complex architecture discussion, switch to GPT-5.6 family models for a quick formatting task, switch to a local Ollama model for editing files that contain proprietary code you don't want leaving your machine. The model dropdown is always accessible in the conversation panel.
Project Management
Save and organize multiple coding projects with persistent configurations. Each project remembers its model settings, file context state, and conversation history. When you switch between projects (your web app, your CLI tool, your library), the full AI context is restored. In the CLI, you'd need to re-add files and re-establish context every session.
Conversation History and Search
Past conversations are saved per project and searchable. If you asked the AI to explain a design pattern three weeks ago, you can find that conversation by searching for keywords. In the CLI, conversation history scrolls off the terminal buffer and is effectively lost unless you manually log it.
Read-Only Server Mode & Remote Visibility
Version 0.77 introduces read-only server mode so you can expose task progress to other devices or colleagues without allowing modifications, useful for demos, pairing visibility, and safer multi-device monitoring while the primary Mac keeps full control.
Live Voice Transcription for Prompts
Voice input matured through summer 2026 releases, including OpenAI live transcription models (gpt-live-transcribe as a default when OpenAI + voice are enabled). Dictate prompts with shortcuts like ALT+V instead of typing long specs.
Extensions, ACP, and Multi-Model Agents
AiderDesk's extension system packs schedulers, skill galleries, destructive-command guards, Copilot bridges, Cursor SDK/proxy experiments, and more. ACP support lets compatible IDEs treat AiderDesk as an agent backend while keeping approvals and tools under AiderDesk's control.
Who Should Use AiderDesk?
1Developer New to AI Pair Programming
A developer has read about AI coding tools but hasn't tried any because they all seem to require terminal fluency. Aider Desk gives them a familiar GUI: click files to add context, type questions in a chat box, review changes in a visual diff, click to commit. Within 30 minutes of installation, they're making their first AI-assisted code changes. The visual feedback removes the uncertainty of 'did the AI actually change the right file?' that makes terminal-based tools intimidating for newcomers.
2Senior Developer Managing Complex Refactoring
A senior developer is refactoring an authentication module that touches 18 files across 4 directories. In the terminal, managing this context means typing 18 file paths and keeping mental track of which files are included. With Aider Desk, they expand the file tree, check the 18 relevant files, and see them listed in the context panel with read/write indicators. They mark the interface files as read-only and the implementation files as writable. The AI refactors across all 18 files, and the diff viewer shows every change organized by file.
3Privacy-Conscious Developer
A developer working on proprietary code wants AI assistance but can't send source code to external APIs. They configure Aider Desk to use a local Ollama model (CodeLlama, DeepSeek Coder). All processing happens on their Mac. They switch to Claude only for non-sensitive tasks like documentation or test generation. The per-conversation model switching lets them enforce their own privacy boundaries without maintaining separate tool configurations.
4Code Reviewer Using AI as a Second Opinion
A developer reviews a teammate's pull request. They check out the PR branch, open Aider Desk, add the modified files to context, and paste the PR description. They ask the AI: 'Review these changes for potential issues: edge cases, missing error handling, performance concerns.' The AI's analysis appears alongside the actual diff in a format that's easy to cross-reference. They use the AI's observations as input for their human code review.
How to Install Aider Desk on Mac
Aider Desk installs via Homebrew cask. It bundles Aider internally or can connect to an existing Aider installation. You need at least one LLM provider configured (API key or local Ollama).
Install via Homebrew
Run: brew install --cask hotovo-aider-desk. This downloads and installs the Aider Desk application to your Applications folder.
Configure an LLM Provider
Launch Aider Desk, go to Settings > Models. Add your API key for at least one provider: Anthropic (Claude), OpenAI (GPT-5.6 family models), Google (Gemini), or configure a local Ollama endpoint. You can add multiple providers and switch between them.
Open a Git Repository
Click Open Project and navigate to a git repository on your Mac. Aider Desk needs a git repo because Aider's workflow is built around git commits: every AI change is a commit that can be reverted.
Add Files and Start Coding
Browse the file tree, click to add relevant files to the AI context, and type your first request in the conversation panel. Review the diff when the AI responds, and approve or reject the changes.
Pro Tips
- • Start with 2-3 files in context and expand as needed: smaller contexts produce more focused results.
- • Mark type definitions and interface files as read-only so the AI understands them without modifying them.
- • If you already have Aider installed via pip, Aider Desk can use your existing installation instead of its bundled version.
- • For local models, install Ollama first (brew install ollama), pull a coding model (ollama pull deepseek-coder), then point Aider Desk at your Ollama endpoint.
Configuration Tips
Optimize File Context for Edit Quality
The single most useful configuration choice is file context. Too few files and the AI doesn't have enough information. Too many files and it loses focus, wastes tokens, and produces less accurate edits. The rule of thumb: include the files being modified plus the files they directly import from. Mark imported files as read-only. For a typical feature addition, this means 3-5 writable files and 2-3 read-only context files.
Choose Models by Task Complexity
Not every task needs the most expensive model. Use Claude Sonnet 5 or GPT-5.6 family models for complex refactoring, architecture changes, and multi-file edits where accuracy matters. Use smaller or local models (DeepSeek Coder, CodeLlama) for formatting fixes, test generation, and documentation: tasks where speed and cost matter more than peak accuracy.
Use Project Templates for Common Workflows
If you frequently work on the same codebase, save project configurations with pre-selected file contexts. A 'frontend' project template might include your component files and CSS modules. A 'backend' template might include your API routes and database models. Switching between templates instantly reconfigures the AI's context for different parts of the codebase.
Review Before Commit, Always
Aider Desk's diff viewer exists for a reason. AI-generated code edits are usually correct but not always. Spending 30 seconds reviewing a diff catches issues that would take 30 minutes to debug later. Pay special attention to edge cases, error handling, and changes to files you didn't explicitly ask the AI to modify.
Alternatives to AiderDesk
Agentic coding tools exploded in 2026. AiderDesk sits on the open/BYO-key side of the map.
Claude Code
Anthropic's agentic CLI/workbench with subscription rate limits; less of a free-form extension host than AiderDesk.
Cursor
Polished AI IDE with Composer models and credit-based pricing; closed product versus AiderDesk's open desktop orchestration.
Codex / ChatGPT desktop
OpenAI's coding agent experiences folded into ChatGPT desktop; great if you already pay OpenAI, less steerable as a self-hosted control plane.
Cline / Roo-style agents
IDE-embedded BYO-key agents; lighter than AiderDesk's full task platform.
Muse Code
Meta's 2026 terminal coding agent entrant; another CLI-first competitor in the budget agent space.
Aider CLI
The classic terminal pair-programmer; AiderDesk wraps and expands that philosophy with GUI tasks, extensions, and multi-agent features.
Pricing
AiderDesk is free to download and use, no license fee, no seat subscription for the desktop shell itself. Costs come from the models you attach: Anthropic Claude API usage, OpenAI GPT-5.6-class APIs, Google Gemini, OpenRouter, local models via Ollama (electricity only), or subscription-backed extensions such as GitHub Copilot where configured. After Anthropic's mid-2026 Agent SDK policy change, the maintainers recommend standard Anthropic API keys for production rather than relying on subscription-bundled Agent SDK credits. Typical spend varies widely, from a few dollars with local/small models to tens or hundreds of dollars monthly with heavy Opus-class cloud usage.
Pros
- ✓Makes Aider's top-tier AI pair programming accessible through a visual interface
- ✓Visual file context management simplifies multi-file editing sessions
- ✓Diff viewer with syntax highlighting provides clear change review before commit
- ✓One-click git revert gives a reliable safety net for any AI change
- ✓Multi-model support with per-conversation switching (Claude, GPT-5.6 family models, Ollama)
- ✓Read-only file support preserves interfaces and types from AI modification
- ✓Project configurations persist between sessions with full conversation history
- ✓Free tool with no subscription: cost is only the LLM API usage
Cons
- ✗GUI adds startup and memory overhead compared to Aider's lightweight CLI
- ✗Dependent on Aider's release cycle for new model support and bug fixes
- ✗Smaller community than standalone AI editors like Cursor or Windsurf
- ✗Less integrated than purpose-built AI IDEs: it's a companion tool, not a replacement editor
- ✗Electron-based, which means higher memory usage than a native macOS app
- ✗Limited to Aider's edit capabilities: can't perform autonomous multi-step tasks like Claude Code
Community & Support
Aider Desk benefits from Aider's active open-source community. Aider's GitHub repository has thousands of stars and an engaged contributor base. The Aider Discord server is one of the more active AI coding communities, with users sharing model configurations, prompt strategies, and workflow patterns that apply directly to Aider Desk. Aider's benchmarking suite (the SWE-bench lite evaluations) is a reference standard for comparing AI coding tool accuracy, which keeps the community focused on measurable quality improvements. Hotovo provides support through their GitHub issues page for Aider Desk-specific bugs and feature requests.
Frequently Asked Questions about AiderDesk
About the Author
Expert Tips for AiderDesk
X users in 2026 describe AiderDesk as a more consistent harness than some experimental editors when running local or open models, while others still bounce to Codex/Claude Code for defaults.
Release notes themselves are the best expert signal: weekly-class shipping, extension ecosystem growth, and explicit guidance away from fragile subscription-SDK shortcuts.
Related Technologies & Concepts
Related Topics
AI Pair Programming Tools
Tools that enable collaborative coding between humans and AI models.
Sources & References
Fact-CheckedLast verified: Aug 9, 2026
Key Verified Facts
- AiderDesk v0.77.0 shipped in early August 2026 with read-only server mode and live transcription updates.[cite-2]
- The desktop app is free; users pay model providers for inference.[cite-1, cite-4]
- macOS arm64 and x64 installers are published on GitHub Releases.[cite-2]
- 1
- 2
- 3
- 4AI coding tools pricing 2026
Accessed Aug 9, 2026
"Positions Aider as open source + API cost model among 2026 tools."
- 5AI coding assistant pricing guide
Accessed Aug 9, 2026
"Market context for seat vs token spend in 2026."
- 6
- 7
Research queries: AiderDesk 0.77; hotovo aider-desk releases; AiderDesk pricing models 2026