New

• Live collaboration (beta) — host or join with a 10-character code from the chat toolbar. Shared agent chat (runs on the host), live typing ghosts, synced agent mode and model, shared in-IDE browser tab, Monaco co-editing with remote carets, guest terminal input on the host shell, and workspace tree sync.
• +50% first-party usage — Free, Pro, Pro+, and Ultra output-token pools for Boongle first-party models are 1.5× previous limits.
• Whop Pro checkout — upgrade to Pro on boongle.dev with account verification, secure card checkout, and automatic plan activation on your Boongle account.
• BoongleBench — GPT-6 Astra, Sol, and Luna on the public benchmark page; full model chart.

Improvements

• Checkout UX — clearer failed-payment messaging on live checkout; sandbox test-card banner only in sandbox mode.
• Agent & chat — collab-aware sidebar and composer; host-only approve/deny for sensitive AI edits.

New

• Boongle 1.8 Pro — first-party model in the picker (Pro+ & Ultra). 1M context, stronger reasoning than 1.8 Flash, with BoongleBench gains on the same harness. Read the launch announcement or full blog post.
• Thinking trace for Pro — reasoning streams into the collapsible Thinking shimmer block (same UX as 1.8 Flash and Mimo), not mixed into the main reply.
• Reasoning effort — Low, Medium, High, and Max in the model picker for Boongle 1.8 Pro (mapped to the provider’s low / high / max API levels).

Improvements

• Identity & stream masking — Pro responses stay branded as Boongle 1.8 Pro in chat, including thinking text.
• Benchmark page — Boongle 1.8 Pro scores on BoongleBench alongside 1.8 Flash.

Fixes

• Chat persistence after updates — your active conversation restores correctly when the IDE restarts or auto-updates, instead of showing a blank chat.

Improvements

• About Boongle IDE — File → Preferences → About Boongle IDE opens a dialog with a short product overview and your installed version.

Improvements

• Chunked file reads — read_file accepts start_line and end_line (1-based, up to 300 lines per call) so agents explore big files without shell workarounds.
• Read labels in chat — tool lines show the line range, e.g. Read App.tsx (L1-200).

Fixes

• Thinking trace in release builds — Boongle 1.8 Flash and other reasoning models now show the Thinking shimmer and collapsible trace in the official installer, matching dev behavior.

Fixes

• Account registration — signup no longer shows a useless {} error; clearer messages for server failures, duplicate emails, and email confirmation.
• Paid plan models — Pro, Pro+, and Ultra subscriptions unlock the correct model catalog again.

Improvements

• Thinking trace UI — clickable “Thinking” shimmer; expands to “Thought” when done; preserves expand state after the reply finishes; fixes missing spaces in streamed reasoning text.
• Mimo V2.5 Pro — enables MiMo deep thinking so internal reasoning streams into the Thinking block instead of the visible reply.

New

• Drag context into chat — drag files or folders from the left explorer and drop them on the chat input to attach them as @ context chips (same as typing @file or @folder).
• Response feedback — after a completed agent turn, thumbs up / thumbs down asks “Was this response helpful?” A toast says thanks; nothing is saved to the cloud yet.
• Agent finished notifications — optional Windows toast + overlay or macOS notification when an agent finishes while Boongle is in the background. Toggle in Boongle IDE Settings.
• Max & Teams pricing — new plans on the website pricing page ($120/mo Max and custom Teams).
• Partners, Terms & Privacy — new legal and partner pages on boongle.dev with full policy content.

Improvements

• New chat composer layout — empty chats show a full-width input at the top; @ mentions and / slash menus open downward so they stay on screen.
• Font picker — search and A→Z / Z→A sort in IDE settings; Comic Sans MS added to the font list.
• Plan usage copy — Boongle IDE Models now says “Includes Boongle 1.8 Flash and Auto”.
• Agent tool streaming — salvages incomplete streamed tool calls, repairs truncated JSON arguments, and surfaces clearer messages when tool calls fail or max agent steps are reached.
• Reasoning models — model reasoning/thinking streams into a collapsed Thinking block instead of the main reply.
• Website redesign — new homepage, downloads, beta, checkout, pricing, changelog shell, expanded FAQ, and UTF-8 symbol fixes across the site.

Fixes

• Model picker scroll — scrolling the model list no longer jumps back up when the cursor passes over items.
• New chats no longer insert a placeholder welcome message row.

New

• Agent dashboard terminal — icon-only toggle in the agents header opens a resizable right-side terminal panel with tabs, close button, and the same shell as the editor. Sessions persist when you switch between Agents and Editor.
• ask_user tool — the agent can pause and ask you a question with styled options, optional custom text input, and accept/decline actions. Answers stream back into the chat as structured blocks.
• DeepSeek model icon — catalog models whose selector or display name contains “deepseek” show the official DeepSeek whale mark in the model picker.

Improvements

• Tab completion ghost text — multi-line AI suggestions now preview the full completion in the editor (not just the first line) before you press Tab to accept.
• Model picker “Fast” label — “Fast” appears only on models that support Fast mode and only when Fast is actually enabled (built-in and cloud catalog models no longer all show it).
• Homepage — refreshed marketing hero and layout on boongle.dev.

Fixes

• Switching to a model without Fast/High toggles no longer clears your global Fast mode preference.
• Tab ghost preview positioning and TypeScript layout types for multi-line completions.

New

• Boongle logo (SVG) — geometric B mark in the title bar, splash screen, sign-in, AI chat, agents view, and empty editor.
• Collapsible chat composer — compact pill when idle; click or hover to expand and start typing immediately.
• Model brand icons — Claude, ChatGPT, Gemini, Xiaomi (Mimo), GLM, and Boongle logos in the model dropdown.

Fixes

• Inline edit — Ctrl+K / Cmd+K no longer collapses multi-line code into a single line; indentation and line breaks are preserved.
• Collapsed chat input shows a text cursor on hover and focuses the textarea when clicked.

New

• Inline edit — press Ctrl+K (or Cmd+K on macOS) in the editor to describe a change and apply it in place.
• Replit Agent — optional Replit MCP tools in chat (enable in Boongle IDE Settings → Replit Agent).
• 80+ UI fonts — pick from a much larger font list under Customization (Google Fonts, loaded on demand).
• Rainbow accent — new accent preset that smoothly cycles hue across buttons, highlights, and the editor theme.

Improvements

• Model picker in chat shows about six models at a time with scroll for the rest.

New

• Agent dashboard overview — welcome home screen with local output token stats, your 3 most recent chats, and top 3 chats by message count.
• Discord Rich Presence — show what you're doing in Boongle IDE on Discord (toggle in Boongle IDE Settings).
• @docs URL mentions — paste a documentation URL in chat for one-off context without adding a permanent docs source.
• Repo agent rules — commit BOONGLE.md or .boongle/rules/ in your project; the agent loads them automatically (like AGENTS.md).

Improvements

• Usage improvements.
• Editor tabs no longer duplicate when switching between the editor and the agent dashboard.
• Smoother sign-in and usage sync when offline or account services are temporarily unavailable.

New

• Command palette — Ctrl+P quick open, Ctrl+Shift+P for commands (file, view, preferences, and more).
• Open Recent — reopen recent project folders from the File menu.
• Search & replace — replace in files with match case, regex, and include/exclude globs.
• Reasoning effort — edit supported models to choose Low, Medium, or High reasoning.
• Web search in chat — see when the agent searches the web and when results are ready.
• /goal slash command in agent mode to start a goal-oriented turn.
• Parallel subagents — multiple subagent tasks can run at once in a single turn.

Improvements

• Higher included usage limits on Free and Pro plans.
• Profile panel shows additional usage credits.
• Monthly usage resets apply correctly when your billing period rolls over.

Fixes

• Chat no longer vanishes when you switch Editor ↔ Agents after sending a message.
• One shared chat instance stays mounted instead of remounting two separate sidebars.
• First message in a new chat is saved immediately, even before a session id existed.

Fixes

• Stopped Auto/Extended from cancelling itself after 6 seconds of thinking, which turned real replies into “No response received.”
• Stopped chat UI state from wiping streamed tokens before they were saved.
• Hung MCP servers no longer block the whole chat request.
• DevTools (F12) works again on Windows so chat issues can be diagnosed.

Fixes

• Fixed additional cases where "No response received" could appear even though the AI had already generated a real response.

Fixes

• Fixed an issue where the AI chat could show "No response received" even though a real response had been generated.
• Improved reliability of usage tracking under heavy load.

Usage

• Usage improvements.

Usage

• Usage improvements.

Usage

• Usage counters stay accurate on IDE restart — manual resets in the dashboard stick.
• Referral panel and bonus API credit / first-party usage rewards.

Chat & agent

• AI todo checklist tool — the agent tracks multi-step tasks in chat with live updates.
• Fixed “No response received” after generation and chat save race conditions.
• Ctrl+. cycles agent modes (Agent → Ask → Plan → Debug).

Models

• Removed the extra hardcoded Auto entry that duplicated cloud catalog models.
• Auto Extended (catalog) and Auto (Old) (legacy Fast/High) are the only Auto family models shown.
• New chats still default to Auto Extended.

Models

• New first-party Auto model (extended context). Legacy Auto is now Auto (Old) with Fast/High toggle.
• Custom models: Reasoning toggle, max tokens, and temperature per slot.
• Add Custom Gemini Model — enter a Google Gemini API key and model ID to use Gemini via OpenAI-compatible endpoint.

Agent dashboard

• Agent Rules modal is larger with bigger text areas for editing rules.

Chat persistence

• Chats save immediately when you send a message — no longer lost if you hide the panel mid-stream.
• Flushes chat to disk when leaving the chat panel or switching Editor ↔ Agents.
• Fixed race conditions that could drop chats from the sidebar list or leave broken ghost entries.
• Safer recovery on startup when reopening a chat that was interrupted.

AI agent

• Removed the idle auto-finalize timer that speculatively ended agent turns.
• The agent no longer stops after a single tool call while more work is pending.
• Busy state now only clears once the turn has actually finished.

Chat file diffs

• File change cards now stretch to full chat width instead of sitting at half width.
• Diff preview auto-sizes height and wraps long lines cleanly.
• Accept / Reject buttons disappear immediately after you accept or reject.
• Fixed path matching so pending changes clear correctly from chat cards.

Chat

• Fixed chat composer still showing Stop / working state after the assistant response finished.
• Fixed a race where Auto model thinking-stall retries could re-enable busy state after a completed reply.
• Added idle detection so visible completed responses auto-finalize within ~1.2s.

Chat UI upgrades

• New Motoko-style chat composer with a cleaner toolbar, send/stop controls, and improved model picker.
• Model dropdown shows cost, speed, and effort on hover — pick the right model without guessing.
• Agent mode selector (Agent, Ask, Plan, Debug) moved into the chat toolbar for quick switching.
• Assistant messages no longer show the star avatar — responses use the full chat width for more readable code and diffs.
• Locked Pro models are clearly grayed out with a lock icon and upgrade tooltip.
• Fixed dropdown menus clipping at the bottom of the IDE and options not being clickable.

Slash commands

Type / in the chat input to open the command picker. Filter as you type.

• Modes — /agent, /ask, /debug, /plan
• Prompts — /test, /fix, /explain, /refactor, /doc
• Context — /file (active editor file), /terminal (recent terminal output)
• Explore — /explore (brief project scan), /deepexplore (deep architecture exploration)

Customization

• New Customization tab in Boongle IDE Settings.
• Accent color presets (orange, blue, purple, green, and more) plus a custom hex picker — applies across the IDE instantly.
• IDE UI font picker with Inter, Roboto, Open Sans, JetBrains Mono, and other Google Fonts (editor font stays separate).

Minimal plugin support

• New .bplugin format — install, enable, and configure plugins from the Extensions view.
• Built-in Supabase and Firebase plugins with sidebar panels and agent tool hooks.
• Marketplace catalog integration — browse and install approved plugins from the website.
• Plugin settings surface in Boongle IDE Settings when a plugin contributes configuration fields.

Boongle Databases

• Create and manage hosted database projects directly inside Boongle IDE.
• Tables browser, SQL editor, and authentication user management.
• Agent tools so the AI can query and manage your Boongle Database during coding sessions.
• Plan-based project limits and API routing fixes for empty project lists.

Agent & models

• Subagents — the main agent can delegate focused tasks; view progress in cards and open a detail lightbox.
• Custom models (paid plans) — add up to two OpenAI-compatible API endpoints in settings.
• MCP server configuration in Boongle IDE Settings for external tool integrations.
• Per-model profiles with descriptions, cost, speed, and effort metadata.

API included credit

• Pro — $20 of API usage (unchanged).
• Pro+ — $40 of API usage (was $60).
• Ultra — $60 of API usage (was $300).
• Usage bars and profile text now reflect the new amounts automatically.

Settings Usage Overview

• New Usage section in Boongle IDE Settings with Boongle IDE Models and Other Models progress bars.
• Shows included plan quota, percent used, and reset date in one place.
• Boongle IDE Models includes Boongle Grok 4.6 and Auto.

macOS Update Notifications

• macOS now checks for new versions on launch.
• When an update is available, a popup asks if you want to go to the downloads page.
• Windows keeps one-click in-app installer updates.

Environment Variable Manager

• Opening .env files shows a choice: edit normally or use the Environment Variable Manager.
• GUI for adding, editing, importing, and saving env vars without raw file syntax.

Improved Profile Viewing GUI

• Wider profile modal (480px) for clearer usage and plan details.
• Expandable usage bars — tap the chevron to read what first-party vs API usage means.
• API usage description shows your plan's included API credit ($20 Pro, $40 Pro+, $60 Ultra).
• Free plan hides API usage expansion; locked state is clearer.
• Log Out button in the profile header for quick sign-out.

Added Agent Safety Settings & Features

• Chat prompts before the agent creates or edits sensitive files (.env, .env.local, settings.json).
• Optional confirmation before reading risky files (off by default).
• Toggle terminal command confirmation on or off in Boongle IDE Settings.
• New Agent safety section in Boongle IDE Settings with all three options.
• Settings gear on risky-file prompts opens Boongle IDE Settings directly.

What’s new

Boongle Grok 4.6 is now available in the model picker. It is based on Grok 4.6 by SpaceXAI, tuned for how people actually work in Boongle: long coding sessions, multi-file edits, terminal use, and agent loops that need to stay coherent instead of drifting.

This is the model we recommend when the task is complex — refactors that span a whole project, tricky bugs, architecture decisions, and writes where “almost right” is not good enough. It is also cost-efficient on first-party usage, especially while the limited-time discount is active.

You do not need a new IDE build for future catalog models. Boongle Grok 4.6 ships through our model catalog, so it appears for everyone as soon as it is enabled. Open the selector, choose Boongle Grok 4.6, and start a chat.

Included in every plan

Boongle Grok 4.6 is a first-party model. It uses your included Boongle usage pool — the same pool as Auto — not Pro API credit. That means it is available on Free, Pro, Pro+, and Ultra. There is no plan lock and no extra checkout step.

Because it is more capable than Auto, it draws usage at a higher rate: 4× Auto per output token. A 40% discount is currently applied, so you are charged 40% less usage than the full 4× rate until we announce that the promo has ended. We will notify you in the IDE when that discount expires.

Free still has a smaller monthly pool, so Grok 4.6 will consume it faster than Auto. If you live in the model for large agents, Pro and above give you a much larger first-party budget. The model itself is the same on every plan — only the size of the pool changes.

How it compares

Auto remains the default. It is fast, cheap, and the right first stop for small edits, tab-style completions, and everyday questions. Boongle Grok 4.6 is the step up when Auto would need several retries, or when you want a single pass that understands the whole task.

Pro API models (Claude Sonnet 5, Gemini 3 Flash, GLM 5.2, GPT 5.4 Nano, Mimo V2.5 Pro, and others) still bill against included API credit. They are excellent specialists. Grok 4.6 is our generalist flagship on the first-party side: one model you can leave selected for hard coding work without burning API credit.

API models require a paid plan with remaining API credit. Boongle Grok 4.6 does not — it only needs remaining first-party usage.

What we optimized for

Stability first. Grok 4.6 is the model we trust to follow tool use, keep file paths honest, and finish an edit instead of narrating one. That matters in an IDE: a flashy answer that skips the patch is worse than a quieter model that actually writes the file.

Complexity second. Multi-step agents, plans you accept before build, and diffs across many files need a model that does not lose the thread. Grok 4.6 is the strongest option we have for that class of work today.

Cost third. Flagship quality on the first-party meter — with a launch discount — is how we want people to try it. Use Auto when you want maximum mileage. Switch to Grok 4.6 when the problem is worth the extra usage.

How to try it

Open Boongle IDE, start a chat, and select Boongle Grok 4.6 in the model menu. It supports image input. Creating and Editing indicators still show the live filename while the model writes.

If you do not see it yet, update to the latest IDE or wait a moment — catalog models load from our servers without a forced app update. The in-app banner “Boongle Grok 4.6” points at the same launch.