Open Source Releases

📹 MoneyPrinterTurbo — Turn a single prompt into HD short videos using LLMs wired into video synthesis pipelines. Fully open-source, targeting automated content creation workflows.

🤖 Claude Code v2.1.158 — Auto mode now works on AWS Bedrock, Google Vertex AI, and Azure Foundry for Opus 4.7/4.8. Opt in with CLAUDE_CODE_ENABLE_AUTO_MODE=1.

🤖 Claude Code v2.1.157 — Plugins in .claude/skills auto-load now — no marketplace required. New claude plugin init <name> scaffolds plugins, plus autocomplete for /plugin arguments.

🛠️ openagentd 1.37.0 — Self-hosted agent framework with streaming chat, tool use, persistent memory, and multi-agent coordination. Built for production agent systems that need stateful, collaborative workflows.

📦 helix-context 0.6.1 — Context management via a “weigh, don’t retrieve” index layer. Could be useful if you’re squeezing more out of context windows in RAG pipelines.

💼 Twenty — Open-source CRM designed with native AI from the ground up. Think Salesforce alternative, but actually extensible and built for embedding ML-driven features.

AI Dev Tools

🛠️ Cline CLI v3.0.15 — Cline Hub lets you monitor connected clients, drive sessions, and stream assistant output via a web app. Adds global AGENTS rules for cross-session agent behavior and plugin-contributed rules.

🧠 claude-mem — Captures everything an agent does, compresses it with AI, and injects relevant context into future sessions. Works across Claude Code, Codex, Gemini, Copilot, OpenCode. Uses Chromadb. Actually tackles the agent state problem instead of ignoring it.

Community Finds

🎨 Taste-Skill — A skill definition that tries to make AI output less generic and more creatively engaging through structured prompt engineering. Could help if your LLM keeps giving you the same bland responses.

Today’s Synthesis

The agent tooling space is converging fast on a shared pain point: state. claude-mem is tackling agent memory loss head-on by capturing, compressing, and re-injecting session context — a pattern that pairs naturally with openagentd , which already supports persistent memory and multi-agent coordination. If you’re building production agents that need to carry knowledge across sessions, combining claude-mem’s compression layer with openagentd’s stateful orchestration could save you from reinventing context plumbing. Meanwhile, Claude Code’s plugin auto-loading signals that the ecosystem is standardizing on local, composable agent extensions — less marketplace lock-in, more control. The takeaway: we’re past the “demo agent” phase. If your tooling doesn’t handle memory, coordination, and extensibility as first-class concerns, you’re building on sand.