Google DeepMind’s DiffusionGemma 26B uses parallel text diffusion to generate text faster and remember more context. This open‑source model can write up to 1,100 tokens per second and supports a 256 K token window.
OpenCrabs 0.3.19 brings native OAuth for Codex, an external embedding API, and a lightweight memory mode for VPS. The release is ideal for developers, researchers, and small businesses that want to run AI agents on their own servers.
A clear guide to open source ai agents, covering what they are, security, architecture, and a simple plan to start today.
Hermes Agent 0.13.0 introduces a Reflective Phase that lets the agent learn from its own actions, a significant token volume increase, and cleaner reasoning markers. These updates make the agent faster, smarter, and easier to integrate into real‑world projects.
OpenCrabs self‑healing agent now supports video vision, auto‑JSON repair, and a new skill system. Learn how this Rust‑based AI can fix itself and run on any channel.
OpenCrabs is a self‑hosted AI agent that learns and fixes itself. Built in Rust, it runs locally, keeps data private, and can auto‑generate new skills.
OpenCrabs AI agent is a self‑hosted bot that learns from its own actions and fixes mistakes without human help. It runs locally, keeping data private while improving over time.
Google’s Gemma 4 is a 31‑billion‑parameter open‑source language model that’s now the third‑ranked open model worldwide. It can run on a single high‑end GPU, supports fine‑tuning, and handles over 50 languages.
Frontman AI tool is an open‑source bridge that lets AI agents interact with a live web app by exposing the DOM, logs, and events as MCP resources.
OpenClaw’s Model Context Protocol (MCP) is a lightweight standard that lets AI agents share data cleanly and reliably. It reduces boilerplate, improves error handling, and enables complex multi‑agent workflows.