Open Source Releases

agent-guardian 1.0.0rc10 — Open-source red teaming toolkit for AI agents, RAG, and MCP servers. If you’re shipping agentic workflows, this gives you a structured way to stress-test them before someone else does 🛡️

openai/whisper — The speech recognition model that quietly became infrastructure. Still the baseline to beat for multilingual STT, and the weak-supervision approach remains a masterclass in scaling data 📄

agent-commerce-protocol-mcp 1.0.4 — Bridges Stripe ACP, Google AP2, and Coinbase x402 so your agents can actually get paid. The plumbing nobody wants to write but everyone needs 💳

chimera-memory 0.8.0 — Local-first reliability ledger for AI coding agents. Persistent, auditable memory that survives context window limits and agent restarts. Finally, an agent that remembers what it did yesterday 🧠

evi-assistant 0.24.0 — Local-first personal AI assistant with CLI, web, and desktop frontends sharing one agent core. Clean reference architecture if you’re building multi-interface local agents 🖥️

xpander-sdk 2.0.344 — BaaS SDK for AI agents handling deployment, tool orchestration, and runtime. The “we’ll run your agents” play — convenient until you need to debug something at 2am ☁️

AI Dev Tools

microsoft/VibeVoice — Microsoft’s entry into open-source frontier voice AI. Another TTS contender — we’ll see if it actually ships or joins the graveyard of “coming soon” demos 🎙️

Today’s Synthesis

If you’re moving agents from demo to production, three pieces from today form a practical stack: chimera-memory gives you persistent, auditable context that survives restarts; agent-guardian lets you red-team the resulting workflows before users do; and agent-commerce-protocol-mcp handles the payment plumbing so agents can actually transact. Wire them together: Chimera logs every tool call and decision, Guardian stresses the full loop (including memory retrieval edge cases), and ACP-MCP settles the invoice when the agent completes a billable task. The integration points are straightforward — Chimera’s ledger is just structured JSONL, Guardian accepts custom attack scenarios via YAML, and ACP-MCP exposes a thin FastAPI wrapper. Start with a single agent that books API credits on your behalf, run Guardian’s injection suite against the memory layer, then add Stripe webhook handling via ACP-MCP. You’ll have a minimally viable autonomous billing agent by Friday — and a test harness that catches the weird failure modes before they hit prod.