Open-Source AI Agent Projects Found Today: June 10, 2026
Today's OpenAgent discovery pass found new or newly active projects across models, agents, memory systems, skills, plugins, tools, and bots.
Today's discovery pass found one useful addition in every OpenAgent category. The theme is clear: agent infrastructure is spreading outward from frameworks into browsers, memory, skills governance, MCP data connectors, observability, edge inference, and voice bots.
Today's additions
| Category | Project | Why it matters |
|---|---|---|
| Models | LiteRT-LM | Google's open-source edge LLM inference framework |
| Agents | BrowserOS | Open-source agentic browser for AI-native browsing workflows |
| Memory systems | OpenLore | Persistent architectural memory for AI coding agents |
| Skills | SkillHub | Self-hosted enterprise registry for agent skills |
| Plugins | MongoDB MCP Server | MCP connector for MongoDB databases and Atlas clusters |
| Tools | OpenLIT | OpenTelemetry-native AI observability and evaluation platform |
| Bots | Dograh | Self-hosted voice AI platform with visual workflows and telephony |
What stood out
The most interesting projects are not trying to replace the whole stack. They each own one layer: model runtime, browser surface, architectural memory, skill governance, database tools, observability, or voice workflow execution.
How to evaluate this batch
Start by matching each project to one workflow. For BrowserOS, test a real browsing task. For OpenLore, index one repository and inspect whether the memory is useful. For MongoDB MCP Server, use a development database first. For OpenLIT, instrument one agent path before adding dashboards. For Dograh, test handoff and telephony behavior before production calls.