AI Product Updates: New Features and Releases

LiteLLM Lens now ships an auto-populated collection of common agent failure modes, beyond only the failure types you define yourself, for building agent self-improvement loops.

osbook launched an agent quest: post and engage, launch a token through the Agentic Launchpad, and scan a contract with Contract Scan, with 0.6 BNB shared among six winners.

Capy 0.4.3 renames Tasks to subagents, supports the full 1.05M GPT-6 context on Codex, and speeds up desktop startup.

Liege Agents built an isolated execution adapter for Virtuals' ACP, added account controls over MCP, and introduced a service catalogue, partial refunds and experimental x402 payments.

AWS MCP Server is now available in six AWS regions including Tokyo, letting AI coding agents call AWS APIs through a single interface with lower latency.

One SDK covers Grok text, voice, image and video models and bundles real-time X search, web search, code execution and remote MCP for building multimodal agents.

Harmonic Agent released a shared liquidity hook for tokenized stocks that blocks fills during trading halts, with some theorems machine-verified in Lean 4.

Silvia, an AI product focused on tax, mortgage and credit cards, says it will launch something new next week and is taking sign-ups for early access.

PineTS runs Pine Script v5/v6 directly with live streaming, multi-timeframe support and custom data sources, plus a CLI for Claude and Codex.

Syra launched its agent launchpad, with the Mimic desk running eight agents that mirror top traders, including creator fee splits and withdrawal rules.

Each dot gets its own cloud computer and keeps working while you are offline, sorting overnight messages by urgency, drafting replies, watching bug reports and migrating legacy APIs.

Agents API now supports launching a browser agent with a single computer-use call, Bedrock managed agents, environment size options, subagent configuration from the dashboard, and better reliability.