Shipped Oct 23, the new submit_feedback tool lets an agent send bug reports, friction reports, and feature requests to the Vercel team with structured context like the tool, task, expected and actual behavior. It requires sign-in and only sends when you ask your agent to share feedback.
Announced Oct 23, the server at <OIM URL /OpsHubWS/mcp lets Claude Code, Claude Desktop, VS Code with Copilot, Cline, and Continue check integration health, build Jira/ServiceNow/Azure DevOps mappings, and retry failed records under the user's own OIM permissions. Delete operations are blocked, MCP changes are tagged in audit logs, and it needs the Professional or Ultimate edition plus a local account or API key, since LDAP and SAML users can't sign in (docs)…
A new AI Quick Start page (announced Oct 23) points Copilot CLI, VS Code, Claude Code, Cursor, and any remote HTTP MCP client at https://ai.orchardcore.net , giving agents tools to search and read the .NET CMS framework's documentation instead of guessing module names and APIs. In Claude Code it's one command: claude mcp add transport http orchardcore https://ai.orchardcore.net (This week in Orchard)…
Announced Oct 22, agents can search Interchange's docs and call any API method in its sandbox to open, fund, and trade accounts, under a Docs Only, Read Only, or Read & Write key you approve in your browser. Each call swaps the agent's token for a 60 second one recorded on the action, and the command tool never retries on its own, so a timeout can't turn into a second trade…
Since Oct 22, the usage metrics API reports customer defined skills, custom agents, MCP servers, slash commands, and plugins by their configured names instead of other or custom , and lists up to 25 per record instead of five. Names come from Copilot CLI, the Copilot app, ACP integrations, and SDK based IDE agents (CLI 1.0.97+ or SDK 1.0.21+), and MCP counts still track connection attempts, not tool calls…
Posted Oct 23, the 428B total/23B active multimodal MoE with 1M context ships with a DSpark draft head and serving recipes for SGLang, vLLM (day 0 images), and TokenSpeed, at roughly 233 GB of MXFP8 weights. It's not Apache or MIT: the MiniMax Community License requires a "Built with MiniMax M3.1 Flash" notice for commercial use and written approval above $20M in yearly revenue (license)…
Spotted something we missed? Start a thread.