As of Oct 14, Harbor can run every eval task in Docker Sandboxes—microVMs that isolate agents like Claude Code, Codex, and Cursor with configurable network policies for fair, reproducible benchmarks. Nightly harbor[sbx] is available now; stable lands Oct 23, with the same tasks runnable locally or on Docker Cloud Sandboxes.
As of Oct 16, AI game engine Summer Engine can be driven by your coding agent through a hosted MCP at mcp.summerengine.com/mcp (assets, store pages, analytics) or a local summer mcp that builds scenes, writes scripts, plays the game, and reads errors. Terminal first agents can skip MCP and use the summer CLI with dry run and json (needs summer engine 4.2.0+)…
As of Oct 15, Vercel Token Pass is in early access for Pro teams: $30 per month buys $80 of shared credit for open weight GLM 5.3 and Kimi K3 through AI Gateway, usable from coding agents or the AI SDK. Zero data retention is on by default; unused credit does not roll over, and requests stop when the credit runs out until the next billing period…
As of Oct 15, Vercel Drives is generally available on Hobby, Pro, and Enterprise: mount a Drive as a directory in a Vercel Sandbox and its files survive after the sandbox stops, so an agent's workspace or file based memory carries over between runs. You can also fork a Drive into independent writable copies so parallel workers start from the same files without touching the original…
As of Oct 15, incident.io packages the skills behind its own Investigations agent (incidents, on call, postmortems, telemetry queries, runbook authoring) with its remote MCP server as one plugin for Claude Code, Codex, Cursor, and Copilot CLI. Install from those marketplaces or from the incident io/skills repo; the MCP server uses OAuth on first use…
NVIDIA released open weights for Nemotron 3.5 Super VL on Oct 15, a Mamba 2/Transformer hybrid MoE (120B total, 12B active) that takes text, images, and video with up to 1M tokens of context for coding, tool use, and document work. BF16 and NVFP4 checkpoints are on Hugging Face under OpenMDW 1.1, with hosted access via OpenRouter and NVIDIA NIM…
Spotted something we missed? Start a thread.