AI news

Polar v1.0.0

Version 1.0.0 of @polar-sh/nextjs moves to the new Polar SDK (@polar-sh/sdk/2026-10) and its standalone functions, so app bundles carry only the API operations each adapter calls. Checkout requests now accept discount_code, which is applied before the redirect or the embedded checkout opens.

Product updatePolarGitHub Releases

Ollama now supports Jev-style decision models

Decision models on Ollama are fast, as requests don’t have to travel over a network. Nimble 9B averaged 91ms per decision in the Pac-Man example below when running locally on an M5 Max.

From the builderOllamaOfficial blog

Documenso v2.19.0

Documenso v2.19.0 introduces a team document analytics dashboard and inbox filters, and documents created from templates can follow naming options. The signing reason can be overridden, and template view tabs give way to a filter pill. PDFs protected by an owner password are now accepted, and action authentication works better for passwordless users.

Product updateDocumensoGitHub Releases

Ollama v0.35.0

Ollama v0.35.0 serves decision models through /v1/systemone; instead of text they return choices, probabilities and scores, which suits ticket triage, model routing and content classification. Nimble from Bespoke Labs and Tev1 from Together AI are the first models available. Settings also open without waiting for model discovery, and MLX downloads no longer hang.

Product updateOllamaGitHub Releases

Easydict v2.24.0

Easydict 2.24.0 adds GitHub Copilot CLI translation with dynamic model selection, saves looked-up words to a local vocabulary notebook and adds pronunciation to Youdao English word lookups. OCR once again recognizes QR code payloads. DeepL now supports Vietnamese. A problem with input translation being cleared has been fixed.

Product updateEasydictGitHub Releases

Ollama v0.40.0

On Apple Silicon, Ollama v0.40.0 automatically runs model architectures that the MLX runtime supports on MLX, with gemma4, qwen3.6 and qwen3.5 among the added models. Decision models and the embeddinggemma-2 embedding model now run on MLX as well.

Product updateOllamaGitHub Releases

Zed v1.21.0

Zed v1.21.0 lets SuperGrok subscribers sign in and use Grok models in the Agent Panel, and a new agent.prevent_idle_sleep setting, on by default, keeps the machine awake while agent threads run. You can now bring your own key for Claude Opus 5.5 and for GPT-6 Astra, Sol and Luna. Syntax-highlighted files and Markdown code blocks draw faster, and a picker for language server commands has been added.

Product updateZedGitHub Releases

See what we're shipping at Latitude.

Latitude is launching Agent Score in early access, enabled per organization, which rates a project from 0 to 100 over a rolling window. The composite weights five dimensions (outcome, reliability, cost, speed and safety) at 35%, 25%, 15%, 15% and 10%, and is withheld when any dimension cannot be measured. A daily sweep picks the shortest whole-week window that meets the session floor, from 7 to 28 days, and published snapshots never change. The score is readable through the REST API, MCP, the TypeScript and Python SDKs, and the CLI.

Product updateLatitudeProduct release notes 1 more update

Rapid-MLX v0.15.0

Rapid-MLX v0.15.0 moves to a new versioned telemetry pipeline that sends only coarse product signals from official builds after disclosure, and raw prompts, responses, file paths and API keys are never included. Users can switch telemetry off in Settings or with rapid-mlx telemetry off at any time, and the old v1 collector is removed. Qwen-Image 2.1 gets dedicated Server and Rapid Mac paths for text-to-image generation and single-image editing. An experimental mimo-v2.6-flash-4bit alias brings Xiaomi MiMo-V2.6 Flash, text-only, to Mac Studio-class machines with at least 192 GB of unified memory.

Product updateRapid-MLXGitHub Releases 1 more update

Dub Letting partners switch groups via application

Existing Dub partners can now apply to join another group in a program through that group's application form. Approval moves the partner to the new group, while a rejection leaves their group unchanged.

Product updateDubProduct release notes

Latitude v7.15.0

Latitude cli 7.15.0 accepts LATITUDE_PROJECT_SLUG or the global --global-project-slug flag to supply the project, so project-scoped commands no longer need --project-slug each time. The generator upgrade adds global flags including --human, --no-retry and --spec. The Linux release binaries now use native-tls with OpenSSL vendored in rather than rustls, adding roughly 6 MB. The macOS and Windows builds are unchanged.

Product updateLatitudeGitHub Releases

Latitude v9.14.0

Latitude python-sdk 9.14.0 adds client.annotations.get, update and delete for managing API-created annotations by their Latitude-generated IDs.

Product updateLatitudeGitHub Releases

Pake v3.17.0

With Pake V3.17.0, a reopened web app goes back to the page you were last on. Apps on Linux X11 stop crashing when a video starts playing, and Windows machines that only have MinGW can build with --windows-toolchain gnu. Codex users can add the official skill from the plugin marketplace.

Product updatePakeGitHub Releases

Easydict v2.23.0

Easydict 2.23.0 lets the Codex service sign in with ChatGPT and install its components when needed, and the Claude Code CLI service can now set its model and reasoning effort. Favorites and history can be cleared in one click, and finished results can be translated back the other way. A 15-second network timeout no longer cuts off long LLM chats, and text selection works in Zen Browser.

Product updateEasydictGitHub Releases

Folo v1.14.0

Folo Desktop v1.14.0 adds an incremental sync engine, so subscriptions, unread counts, collections, lists and timelines follow the server's change log rather than reloading full snapshots. Marking items as read or unread, marking everything read and starring now take effect right away, with changes queued and retried while you are offline. API calls, updates, downloads and other traffic now run through Chromium's network stack and honour system proxies, PAC files and SOCKS, and a temporarily unreachable API no longer signs you out locally.

Product updateFoloGitHub Releases

Cherry Studio v2.1.0

Cherry Studio v2.1.0 introduces System Doctor, which uses AI to check installation, permissions, storage, providers, MCP and more, then walks you through fixes. Agents and chat assistants can drive the built-in browser, and a new office-transform skill pulls worksheet ranges, document paragraphs, PDF pages and slide shapes out into new files. Deleted topics, agents, assistants and files now go to a recycle bin first. From this release on, macOS 13 (Ventura) is the oldest supported version.

Product updateCherry StudioGitHub Releases