... noticed GPT-6 Astra was good at identifying when it had finished a phase and no longer needed the raw tool output, so I built a Codex patch that lets it request compaction itself. The model writes a checkpoint with the relevant decisions, constraints and next steps. The patch runs native compaction

... version selection and rollback; vendor scripts install the current vendor release. OpenCode remains npm-based. Antigravity status integration saves and restores the CLI s original status line. Upgrading from v0.8.7 also brings the v0.8.8 storage migration panel and more compact allowance rows. Release notes and downloads MIT-licensed. I maintain the project; text prepared with AI assistance. Cover: AI-generated concept illustration, not a screenshot. Comment from Bradely Charlesbois(@bradely_charlesbois): @rakesh_gupta20  The live activity status sounds help ...

... restoration checks that the rollout file still exists. Completion uses the main-thread Stop hook to avoid confusing side-thread activity with the end of the task. Event detail size and retained event count are bounded. Existing databases are compacted on the first upgraded launch, which may take a few extra seconds. Try it Requirements: Windows 10 build 17763+ and WebView2 based on Chromium 111+. Download an EXE/MSI installer, or run npx termexo@latest with Node.js 18.18+. Agent CLIs ...

... Previous and Next controls, Play, Pause, Stop, seeking, volume control and automatic playback of the next track. The player can work with the current playlist context and includes direct navigation through the Music.app playlist hierarchy. There is also a compact Mini Player and a Classic interface inspired by traditional hi fi hardware. Playlist Compare has been improved significantly. Changes made manually in Music.app are now synchronized more reliably. The app can detect when a track is replaced in a playlist ...

... There's a moment in every long agent session that almost nobody notices: compaction. When the conversation approaches the context limit, the harness has a model summarize everything so far and swaps the summary in for the real thing. Your scrollback still shows the full conversation. The model's world is now the summary. This explains one of the most confusing agent behaviors: hours into a session, it stops honoring a decision you both settled early. You scroll ... ... attached deserves its own file on disk, where the summary can't reach it (the agent just has to re-read it). The conversation itself should hold nothing you can't afford to lose. And one habit: when the compaction

... original hosts. It s a real backend, not a CLI: FastAPI + an arq worker + a LangGraph pipeline w/ a React PWA. One command (./launch) brings the stack up. Jobs pause at each ML stage so u can pick local vs hosted, model size, length & voices. I keep making it better, adding lots of features to it (searching podcast contents is the most interesting one I can't wait for). I made it opensource(MIT), rn. U can find ... ... machine (self-hosted), so u can use ur own GPU or BYOK to offload the heavy jobs. GitHub:  https://github.com/behradkhodaya... The engine is solid & this is still WIP (it works fine tho & I've started compacting

... back. Smaller things you will notice Update notices in the app. When a new version is ready, a small card appears in the window. See what's new and install when you want, or dismiss it for that version. Alternate Dock icons. Pro users can pick between the classic tile and two new duck icons in Settings, under Appearance. The Dock, the menu bar and the About screen all follow. Pinned tabs as chips. A pinned tab shrinks ... ... compact

... symbols out in a grid on a plate beside the card, unfolding its folder first. Drafts on the board. Click-out asks Save / Save as draft / Discard; drafts stay yours and are never handed to agents. "+ Task" starts blank. Compact density (View Density) and a Changed-files section with one-click diffs. Agent channel Agents get names you can say aloud (Claude 1, Codex 1) and a shared room: they post which files they are taking, mark them ... ... every design decision for your review so you can agree on a second time (or stop your agent beforehand if you prefer a more hands-on approach). This already goes in the direction of flagging the decision point between alternative

... advertised limits. Coding sessions are context-rot factories every git diff, test run, and stack trace turns into dead sediment competing for attention. A paper from June measured agents with safety policies: fresh in context = 0% violations. After auto-compaction summarized it away = 38% violations. Not disobedience, just amnesia. What actually works: Put anything that must stay true in a file the agent reloads every session ( AGENTS.md , CLAUDE.md , whatever your tool reads). Stuff you say in chat at turn ... ... unlucky summarization away from gone. Files survive. Chat doesn't. Several small sessions beat one epic. A fresh session with a written handoff beats a long one with a silent auto-compact

... other displays Every screen can have its own island, or one that follows your pointer. External displays without a notch draw a virtual one sized to their menu bar. Notifications got quieter Toasts never expand the island anymore. The compact pill stretches a little, shows what happened, and settles back on its own. Also in this release: The calendar card pages through today's agenda, settings were rebuilt as searchable sidebar panes with Liquid Glass on macOS 26, and full ...

... Analyzed a day of token logs across an autonomous coding agent setup running on internal codebases. The raw count: 769M input tokens vs 7.4M output tokens (~104:1) . Because long agent runs re-read session history (files, AST diffs, test outputs) every turn, input costs accounted for ~95% of total spend. Optimizing output length turns out to be looking at the wrong variable. Three things actually saved us money: Routing: Shifted non-interactive workloads (evals, background analysis) to batch/flex ... ... with zero code logic changes. Cache Discipline: Kept system prompts strictly byte-stable (no top-level timestamps). Achieved a 94.9% prompt cache share, driving input costs from $10/M down to ~$1.46/M blended. Context Compression: Built a pipeline sending compact

... users can review and correct the result before saving it. They can change the food name, portion, preparation method, quantity, or ingredients. When confidence is low, FitBin can request another picture, ask a short clarification question, or provide likely alternatives. In addition to taking pictures, users can search for predefined foods by name. Search results can include common foods, branded products, restaurant items, recent meals, favorite foods, and previously saved combinations. This allows FitBin to remain useful in situations where ... ... information For example, FitBin may explain that a meal provides enough protein for the user s muscle-building goal but contains more sodium than their remaining daily target. It may recommend a smaller portion or suggest a lower-sodium alternative

... across toolbar buttons for easier window moving; UI Improvements: - Active pane indicator : clearer active pane marker in the status bar when split view is on; - Resizable info panel : drag to resize the info panel width and the preview/details split; - Compact info panel : denser property layout in the info panel; - Context menu actions : `Edit card` is shown as an action button, with smaller action buttons overall; - Navigator styling : improved adaptive layout, tab active-state styles in split view, and command palette ...

... active delivery trucks and log records */ public function index() { // Use eager loading to prevent N+1 query performance lag $activeLogs = TruckLog::with('driver') ->whereDate('created_at', today()) ->orderBy('created_at', 'desc') ->paginate(15); return view('dashboard.fleet', compact('activeLogs')); } } In the Blade view, I mapped these variables directly to the pre-styled table components inside the Arcone template. The transition of static dummy data to actual database records worked nicely, and the table rendered cleanly with Bootstrap ...

... agent can even understand what needs to be tested. We considered Markdown first. Markdown is agent-friendly, reviewable, and easy to version. But for UI test cases, Markdown has a serious limitation: it does not give you a compact global view. In practice, we wanted to see many test cases at once: - case id - version - smoke flag - status - precondition - steps - expectation - comments - latest execution result That is why we chose XLSX. A spreadsheet can show dozens of test cases ... ... worker. For UI testing, a remote headed browser is not just a convenience. It changes the debugging loop. Instead of asking "what did the script think happened?", you can ask "what did the user actually see?" ### Harness Code vs

... updates commit when you release, so boards stay responsive with lots of notes. New stickies are a bit larger by default. Overlapping notes come to the front when you select or drag them. Cleaner sidebar Tag filter is a compact `#` button (no full-width dropdown). New note / board / mind map actions are icon-only `+` buttons with hover tooltips. Fixes - Spaces work properly in note titles again - Checklist Enter and checkbox toggle behave correctly Everything stays local-first in your browser ...

... caching that recognizes when "last 7 days" is really asking for the same data) Way easier member invites: one dialog, roles and groups in the same step, and bad email addresses no longer block the batch Long chats auto-compact instead of hitting a dead end Suggestions launch: https://www.producthunt.com/prod... Full changelog: https://www.basedash.com/changel... ...

... improvements had not been merged Instead of building something completely new, I decided to maintain an enhanced fork. Some of the changes include: bounded, configurable storage that doesn't grow without limits openwolf doctor for automatic maintenance and compaction BM25-powered memory recall with citations an MCP server so the project memory is also available from Claude Desktop and other MCP clients support for additional coding agents such as Codex CLI, Gemini CLI and OpenCode numerous bug fixes, security improvements ...

... Effective input cost/day $2.90 $0.62 Output cost/day (unchanged) $7.50 $7.50 Daily total $10.40 $8.12 Wait, that is only $68/month savings on raw math. Where does the $200 come from? The bigger savings: I stay in sessions longer without hitting compaction. Before compression, long sessions hit compaction

... stay visible even mid-sync, so you always know what's happening. Lighter on your Mac the on-device AI now releases memory when it's idle, keeping Quartz leaner in the background. Email text is a touch more compact, making longer messages easier to scan. Fixed Your signature no longer disappears when collapsing a quoted Apple Mail reply. Comment from Marie Saxon(@marie_saxon): Honestly, search on Mac, especially when it comes to mail, has been painfully ...

... longer stuck with the default and can match a chart to your brand, your dashboard, or whatever convention your team already uses. Numeric charts also get a new Show full numbers toggle. By default Basedash keeps big values compact (like 1.2M), but when you need the exact figure you can flip a chart to show the full number in the primary value and data labels while keeping the axis labels compact

... chromewebstore.google.com/detail/page-jumper/jhmamddnkdlnafeddaioinhodlmcigjn A compact page navigator with progress, headings, saved positions, and quick jumps. Navigate long webpages without losing your place. Page Jumper adds a compact

... tomorrow's weather is ready before the day is over Precipitation alerts: practical reminders when rain may affect your plans Sunscreen alerts: high-UV reminders when sun protection may be useful A new system-small daily widget for a compact look at upcoming daily conditions Each notification type can be turned on or off individually, so you can keep only the ones that fit your routine. The new daily widget joins the existing Weather mini widget family, including current weather ...

... China mirror source choices for Startkit installs, plus an optional shell PATH setup switch. - Reworked onboarding into a simpler step-by-step flow for coding agents, messaging apps, remote access, install, and final configuration. - Refreshed Runtime Status with compact summary cards, branded service icons, status tooltips, and focused detail dialogs. - Added /switch agent profile as a shortcut for /switch host agent profile from IM channels such as Feishu. - Hardened local auth and session handling with stricter local boundaries, rate-limited ...

... Codex / Claude-style launches - New provider profiles: xAI, NVIDIA NIM, and Mimo - Pi profile launches and direct Pi session support - Agent enable controls, bridge proxy controls, and an in-app release update indicator - Web Chat now surfaces context-compaction notices more clearly If you use multiple coding agents or switch providers often, this version should remove a lot of config friction. Release: https://github.com/jazzenchen/Vi... Feedback is very welcome, especially on the new launch/profile flow. Comment from Jazzen Chen(@jazzenchen ...