Friendly release notes for every update

What you would notice

• After a quick action finishes, the side panel refreshes the Writing Assistant session on its own, so the new request and its answer appear without reopening the panel.
• Long sessions now open on your most recent messages instead of the oldest ones, so new quick action results are never hidden behind older history.
• The refresh is silent: the text you are typing in the chat box and your pending attachments stay exactly where they were.

What you would notice

• Choose "Native Messaging" under Settings → Advanced → Communication Mode and every request, including streamed answers, goes through the native host.
• Replies still arrive word by word, and the Stop button cancels the answer in Native Messaging mode too.
• You can switch back to "HTTP Server" at any time; the native connection closes cleanly once pending requests finish.

What you would notice

• Assistant replies now render as Markdown, so headings, lists, tables, and code blocks look the way they were meant to.
• The chat input grows automatically as you type longer prompts, so you can review everything before sending.

What you would notice

• Quick actions (like translating or tweaking text) now resolve the fastest available model dynamically instead of relying on a fixed list, so lightweight tasks stay fast.
• The extension now caches your available models locally and refreshes them about once a day, falling back gracefully to the last known list (or a built-in one) if you are offline.
• Added the new MAI-Code-1-Flash model, along with support for turning reasoning off entirely ('none') on models like GPT-5.4 Mini.

What you would notice

• Dependencies and security overrides were updated to mitigate a large batch of known vulnerabilities.

What you would notice

• Added baseline model constants and default quick action models, paving the way for the dynamic model selection introduced in a later release.
• The default model for new sessions was migrated from the deprecated GPT-4.1 to GPT-5 Mini.

What you would notice

• A new model switch flow makes it easier to change how the assistant responds without leaving the browser workflow.
• Supported models can now use reasoning effort controls, so you can choose between faster answers and deeper thinking.
• The extension became more reliable overall thanks to fixes across hooks, components, utilities, and entry points.

What you would notice

• Chat now handles image attachments more smoothly, including the pre-upload flow behind the scenes.
• The side panel and chat experience now explain image-related issues more clearly when something goes wrong.
• Screenshot quality was improved so visual context is more useful when you ask for help.

What you would notice

• Suggested text can now be inserted at the cursor instead of overwriting the whole draft.
• Session recovery became more resilient when the backend temporarily loses track of an active chat.
• The chat flow was guarded against accidental model mismatches once a conversation had already started.

What you would notice

• A configurable backend URL made it easier to connect the extension to custom or relocated backend instances.
• The release also helped more advanced users work with flexible local setups such as Docker-based backend environments.

What you would notice

• Model search and pricing tiers made it faster to compare options before starting a session.
• Account status became easier to understand from the extension UI.
• Firefox behavior became smoother with better browser compatibility and side panel handling.

What you would notice

• This release mainly prepared the platform for upcoming account, model, and session improvements.
• No major new extension workflow was highlighted in the published release notes.

What you would notice

• Message history now returns the most recent page of a session first, so long chats and the Writing Assistant session always show what just happened.

What you would notice

• The backend package now includes a Native Messaging host that Chrome and Chromium can launch directly, with streaming replies and cancellation working just like over HTTP.
• On Windows, the host installer now registers itself for Chrome and Chromium automatically, and removes those entries again when you uninstall it.
• Native Messaging works alongside the regular HTTP server, so you can switch between the two from the extension settings at any time.

What you would notice

• Running npx devmentorai-server in PowerShell or Command Prompt now starts the backend correctly, with no WSL required.

What you would notice

• The backend now validates that a requested model is actually available to your Copilot account before creating or switching a session, instead of failing later with an unclear error.
• Sessions and model switches now accept an explicit 'none' reasoning effort, so models like GPT-5.4 Mini that support disabling reasoning can do so.

What you would notice

• The backend now resolves the Copilot CLI path more robustly using platform-specific packages, fixing startup failures in some pnpm-based setups.
• Adjusted a dependency override so a required security library no longer gets silently downgraded.

What you would notice

• Dependencies and security overrides were updated to mitigate a large batch of known vulnerabilities.

What you would notice

• The backend can now authenticate using a GITHUB_TOKEN or COPILOT_TOKEN environment variable, useful for headless or server setups.
• The default model for new sessions was migrated from the deprecated GPT-4.1 to GPT-5 Mini.

What you would notice

• Session handling was improved so model switches and reasoning settings behave more smoothly behind the scenes.
• Compatibility work keeps the backend aligned with newer GitHub Copilot protocol and runtime expectations.
• CLI and service updates make the local engine that powers the browser experience more stable day to day.

What you would notice

• The Copilot SDK was updated to support protocol version 3.
• Runtime compatibility was improved so backend sessions behave more reliably in newer environments.

What you would notice

• The backend bundled the Copilot SDK to avoid strict ESM module resolution problems in Node 20+.
• That reduced setup friction and made local startup more dependable.

What you would notice

• No major end-user backend feature was highlighted in the published release notes for this version.
• This release mainly coincided with platform stabilization while newer capabilities rolled out across the product.

What you would notice

• Support was added for reading public web content during assisted workflows.
• Image pre-upload support was added for richer chat requests.
• Permission handling was updated for newer Copilot SDK behavior.
• Docker Compose support made local backend setup easier for users who prefer containerized workflows.

What you would notice

• Account routes and model-management capabilities were expanded.
• CLI diagnostics and backend error handling were improved so problems are easier to understand.