Latest Open-Source AI Agent Projects to Watch in June 2026
A short editor's list of Webwright, AgentMemory, Agent Browser Protocol, Garden Skills, MLflow, Tiledesk, and FinGPT.
The most useful agent projects in June 2026 are not all agent frameworks. The stronger pattern is a stack: browser control, persistent memory, reusable skills, evaluation infrastructure, and domain models.
This month, start with Webwright for long-horizon browser tasks, AgentMemory for coding-agent memory, Agent Browser Protocol for deterministic browser tools, Garden Skills for reusable workflows, MLflow for AI engineering, Tiledesk for support bots, and FinGPT for finance-oriented LLM work.
Quick shortlist
| Project | Category | Why it is worth watching |
|---|---|---|
| Webwright | Browser agents | Microsoft-backed framework for long web tasks |
| AgentMemory | Memory | Persistent context for coding agents |
| Agent Browser Protocol | Plugins | Browser control as a deterministic tool surface |
| Garden Skills | Skills | Reusable instructions for design and creative work |
| MLflow | Tools | Experiment tracking, evaluation, and production AI workflows |
| Tiledesk | Bots | Support bots with human escalation |
| FinGPT | Models | Open financial LLM research and domain adaptation |
What changed
The agent ecosystem is becoming less framework-centric. A serious agent stack now needs controlled tools, durable context, repeatable tests, and a workflow for humans to take over when automation is unsure.
How to evaluate them
Pick one representative workflow, run it with logging enabled, and write down failure modes before adding more tools. For browser agents, measure task completion and recovery. For memory systems, compare session two against session one. For support bots, test escalation before testing model creativity.