LongCat 2.5: API pricing, visual inputs and deployment limits
What developers should check before adopting Meituan’s latest model.
AI Agents
Gemini API CLI gives coding agents a separate route to Gemini models
A practical guide to choosing the right Gemini command-line tool.
MiniMax Code Mini Apps: local workflows and permission limits
A practical guide to MiniMax Code Mini Apps, including local desktop access, persistent data, customization and the limits of host permissions.
Factory Droid 1.0 explained: tool workflows, pricing and permissions
A practical look at Factory’s October 29 harness update, with current plan prices, approval boundaries and the distinction between the announcement and today’s version.
Vercel agent-desktop: platform limits and verification checks
A practical guide to agent-desktop compatibility, accessibility limits, action verification and sensitive desktop evidence.
DeepSeek superagent prototype separates task permission from execution
DeepSeek’s research prototype shows how coding agents admit work, run separate sessions and request review. Its execution permissions still need careful scrutiny.
OpenAI Evals migration: what to check before November 30
A practical checklist for rebuilding evaluations, preserving prompt versions and validating exported agent workflows.
GitHub’s Microsoft Teams app reaches GA with shared coding sessions
GitHub’s Teams app is generally available. See what changed, how shared sessions work, and which repository permissions and review limits still apply.
Antigravity adds a Plugin Marketplace for App and CLI
Google’s new catalog bundles MCP tools, skills and agents. Here is what to inspect before installing, and why setup differs across surfaces.
Google sets Jules wind-down dates and publishes export guide
Jules becomes read-only after November 11 and winds down December 31. Here is what to preserve and what the migration guidance does not establish.
ServiceNow expands OpenAI deal, with integrations still ahead
The plugin is available. Larger integrations remain planned, with licensing checks required.
GitHub Copilot Memory goes GA as Dreaming enters preview
Memory and Dreaming have different availability, controls and review boundaries. Background runs can consume credits without producing changes.
GitHub Agentic Workflows reaches GA with Actions-based automation
Markdown-defined repository automation is generally available, but Actions compute and AI inference still need separate budgets.
GitHub open-sources Replica for prototypes built with your design system
Replica brings a design-guided canvas to Copilot. Local prototypes need no Hub, while shared publishing adds hosting and access requirements.
Copilot Projects and shared automations give team work a new home
Projects organize ongoing work. Shared automations add repository ownership, with separate access and billing limits.
Muse reaches Singapore with local connectors rolling out gradually
Meta opens Muse on mobile and web in Singapore. Partner integrations remain staged, so compare completed tasks rather than assuming every transaction is supported.
GitHub lets Copilot SDK services use an organization identity
GitHub previews organization-owned Copilot SDK authentication. App permissions, token renewal and billing attribution need separate attention before services run unattended.
Liquid AI releases LFM3 for local agents and multimodal apps
Compare LFM3's six small models, official demos, license limits and local runtimes, including the minimum build needed for NVFP4 output.
IBM gives AI agents identities, with new discovery in preview
IBM separates available identity, credential and evaluation tools from new discovery and containment previews. Here is what a bounded workflow should prove.
NVIDIA and Meta expand Muse partnership beyond GPUs
Meta plans more NVIDIA infrastructure for Muse. The deal connects model inference, agent tool use and a possible separate enforcement layer, not a new consumer release.
Oracle plans to bring finance, HR and supply-chain work into ChatGPT
Oracle AI Workspace is coming soon. Its published workflows show how ChatGPT could coordinate delayed orders, month-end close and employee onboarding across Oracle systems.
Open Responses joins AAIF with a new governance charter
Open Responses has joined AAIF. Its new charter changes stewardship of the API specification, while provider compatibility still needs careful testing.
Pictory 3 adds an AI video agent with review checkpoints
Pictory 3 moves from a brief to a reviewable video draft. Here is how the workflow, editor controls, rollout and credit limits affect creators.
Step 5 Preview open weights explained for deployment teams
Compare Step 5 Preview checkpoint sizes, GPU configurations, preview serving support and commercial license conditions before planning a deployment.