Open Source AI Agents Worth Trying in 2026 - Desktop, Browser, and Code
The open source AI agent ecosystem has exploded. Here are the ones worth your time, organized by what they actually do.
Desktop Automation
Fazm - Native macOS agent. Voice-controlled, uses accessibility APIs for reliable UI control. Supports Claude and Ollama. MIT licensed. Best for: automating multi-app workflows on your Mac without writing code.
Browser Automation
browser-use - Python library for AI-controlled browser automation. Uses Playwright under the hood. Good for web scraping and form-filling tasks that do not require native app control.
OpenAI computer-use - Reference implementation of computer use through screenshots. Cloud-based, sends screen captures for processing. Good for understanding the screenshot-based approach.
Code Agents
Claude Code - Anthropic's CLI coding agent. Excellent for writing and modifying code but operates in a terminal, not on your desktop. Pair it with a desktop agent for full workflow coverage.
How to Choose
The decision tree is simple:
- Need to control native Mac apps? Fazm - accessibility APIs work across all apps.
- Need browser-only automation? browser-use - simpler setup for web tasks.
- Need to write code? Claude Code - best coding agent available.
- Need all three? Fazm + Claude Code - they complement each other.
Model Routing
One underappreciated feature: being able to route different tasks to different models. Simple tasks (click this button, fill this field) work fine with local 7B models. Complex tasks (plan a multi-step workflow, draft a nuanced email) benefit from Claude.
Fazm supports both Ollama and Claude, so you can route based on task complexity.
Fazm is open source on GitHub. Discussed in r/AI_Agents.
- Open-Source AI Agents You Can Run Locally on Your Mac in 2026
- Fazm - Open Source Voice-Controlled AI Agent for macOS
- n8n Alternative: When Visual Workflows Cannot Reach Your Desktop
Comments (••)
Leave a comment to see what others are saying.Public and anonymous. No signup.