pocket/grammer keeps getting better on Mac and iPhone: • Mac: try before you buy — pocket/grammer for Mac now includes a free hourly demo: every hour you get a 3-minute window to connect your IDE to the on-device server, no purchase needed. A live countdown on the home screen shows your next window. Unlock unlimited server connect anytime with the new Mac purchase. • One purchase, every device — buying the unlock on iPhone now also unlocks the Mac app (and vice versa) under the same Apple ID. • Text size setting (Mac) — new − / + control in Settings scales text across all screens, from 80% to 160%. • Fix: cloud models now start the Agent — assigned a BYOK cloud model (OpenAI-compatible, OpenRouter, Anthropic, Google, or custom provider) to Plan/Code/Chat and tapped a mode, only to stay stuck on "Pick a mode to load a model"? Fixed — cloud-only setups now start the Agent directly, no local model download required. Plus stability improvements and smaller fixes throughout.

Agent improvements The agent can now run JavaScript in a secure on-device sandbox to verify its own code, and check files for syntax errors — so it catches and fixes its own mistakes before showing you results. Smarter, more reliable tool calling: fewer failed or repeated edits, better recovery when a command can't run, and automatic self-review of each plan step with automatic repair. File deletions now ask for confirmation, and the agent no longer repeats identical writes. Code editor & workspace New AI inline completions: pause while typing and the on-device model suggests the rest of the line (toggle in the editor toolbar). Replace-in-project: find and replace across all project files, with full undo. Project shell: run ls, cat, grep, sed and friends against your project from the new SHELL tab in the console panel. Preview reload button, plus clearer error messages when a preview can't load. Chat New chat role: pick a dedicated model for conversations, separate from Plan and Code (Models tab). Chat answers that get cut off can now be continued with one tap. Enter submits on Mac; the attachment picker is Mac-friendly. Mac (Catalyst) Text is larger by default, with a new Text Size setting (Default/Large/Larger/Largest) in Settings. Settings now shows your actual Mac model and macOS version. Desktop features are included on Mac without an unlock purchase; the GitHub section and workspace GitHub sheet now open correctly. The Settings button in the header is now labeled. Stability & fixes Fixed the preview showing a blank white page instead of your web page. Fixed the on-device server rejecting connections from coding IDEs (Cursor, Zed, Continue) and accepting model names in any form. General reliability improvements across agent runs, model loading, and project sync.

- pocket/grammer is now a universal app! Use on your iPad or Silicon Mac - Reset desktop client trial with every new update - Full push/pull Github support - Import models already downloaded on your device - Improved Planner/Coder handoff - Distinct Chat Mode - On-device OCR for PDF files

- Bring Your Own Key (BYOK): Connect any OpenAI-compatible API key for cloud-powered coding sessions alongside on-device models. Keys are stored securely on your device. - Supercharged Workspace: Drag-and-drop file moving, rename and copy, project-wide search, and a diagnostics bar to catch issues fast. - Smarter Plan → Code pipeline: The agent now asks clarifying questions, validates its plan before writing code, and executes step-by-step with per-step review and undo. - Improved undo & history: Stronger step history across the Agent and Workspace, so every AI edit can be reverted. - Refreshed desktop experience: Pair your phone with your computer for a full desktop coding UI powered by your device's models.

- Small models struggle with multi step tasks, so we’ve split the work between two modes: Plan and Code - Two different modes, two different models. Switch between them on the fly - Plan mode asks questions about what you want to accomplish, writes a plan and hands-off to the coder - Coder only needs to worry about implementation

- Refined Agent behavior - Improved UI - Improved model loading - Updated recommended models - Unlocked more CPU threads - Offloaded some work to GPU - New app icon!