2026
-
00:00
- Fixed calendar block editing, which was broken in a recent regression
02:47
- Fixed an error that could occur when using the 2nd Opinion dropdown
- Fixed a bug where usernames with mixed capitalization would see errors applying chat blocks to notes
15:43
- AI-highlighted text (underlines) now stays intact when you save a note, regression from feb 17th
17:14
- Fixed a regression error that could occur when applying all from a chat to notes, ie the notes wouldn't show up until a refresh.
- Fixed tasks so it more reliably shows the diff automatically
19:21
- Fixed List button disappearing on calendar view
-
21:45
- Backlinks now show ai links and process them after editing
- Backlink editing reliability and ui is improved
- Prompt pages optimized for speed and fewer database queries
- Adjusted audio recording silence trimming for slight improvement ie less hallucinations.
- Changelog page refactored
-
06:01
- Audio transcriptions are now more accurate — silence is trimmed before processing, which prevents the transcriber from hallucinating words that weren't said
- Fixed a bug where audio meetings could create duplicate note blocks
- Minor spacing and layout improvements on datatype pages
Ok how does typing feel. I guess it’s fine.
-
17:48
- Fixed positioning of the unread notifications badge so it displays correctly on all devices
-
17:58
- You'll now get push notifications when an agent replies to your chat, even if you're not in the app
- Unread agent replies are now highlighted so you can easily spot new messages
- Contact sections in task agents are reordered for easier scanning
- Fixed a display glitch where audio blocks would jump around after loading
-
Initial foundation for grugnotes cli laid. Nothing user facing yet! Coming soon.
-
05:28
- Telegram replies now stream in real-time instead of arriving all at once
- Landing page copy and layout improvements
- Fixed an issue where email-based agents could cause repeated errors if the agent was deleted
- Agent access now properly activates and deactivates with your subscription
06:11
- Telegram bot no longer shows raw reasoning tags while streaming responses
07:12
- When you send a note via Telegram that doesn't need a reply, Grug now reacts with an emoji instead of staying silent — so you know your message was received.
-
Massive updates deployed. Monitoring for regressions over the next few days. Docs and videos coming soon on the major features. Also a lot of infrastructure changes to set groundwork for further improvements.
Lists — New "List" prompt mode for managing ordered collections with a full table view:
- Inline cell editing, search, sort, filter, pagination, and column visibility
- Drag-to-reorder rows and columns
- Reference cells with typeahead linking between lists (stored as [[wikilinks]])
- Ref cell popovers for viewing/editing linked item attributes
- Boolean cells render as auto-saving checkboxes
- Adaptive wide layout for tables with many columns
- Lists section in the sidebar
- "Tracker" auto-detected as list mode for new prompts
Wikilinks & References — [[prompt/block]] wikilink support with automatic path sync when prompts are renamed. Block text references are rewritten on rename too.
Entity Merge — Merging duplicate entities now rewrites references across your notes and handles conflicts gracefully.
AI Backlinks — Incremental hash-based diffing replaces the old delete-all/recreate approach. Backlinks now have a confidence level that locks on confirm.
Sage Model — New feature release powered by Claude Sonnet 4.6. Ungated for all users.
Gemini — Upgraded to 3.1 Pro.
2nd Opinion — Model selection dropdown lets you pick which model to use.
Search — Switched off Gemini flash 2 models to Groq/Gemini Flash 3. Smarter results interpretation.
Date Extraction — Improved accuracy for relative weeks, appointment patterns, and HTML-escaped text.
Prompt Attributes — Drag-and-drop sorting, default values, and descriptions for prompt attributes.
Autotitle — Blocks can now be auto-titled.
Collapse/Expand — Toggle to collapse or expand all blocks on a page. List blocks default to collapsed.
URL Linkification — Fixed streamed and markdown-wrapped URLs not becoming clickable links.
Enter to Submit — Input-style submit controls now respond to Enter key.
Fixes
- Chat textarea intermittently hidden after focus
- Month-boundary date prefilter off-by-one
- Scrollbar flicker on prompt header
- Layout flash on page load
- Stale list table after browser back
- "Past" header missing when recent days are empty
- Collapse arrow unreachable on prompt/block pages
- Code block copy button hidden in some cases
- Blocks created under wrong prompt
- Stale static assets served from home cache after deploy
- CSRF error on ASR token count
- Duplicate loading spinners on Publish/Pin buttons
- Sidebar caches not refreshing after async emoji generation
- Stale flash messages on tasks page
- Focus prompt checkbox removed from prompt settings
-
Redesigned block dropdown menu - Cleaner single-column layout with new Change row for quickly editing a block's prompt, date, or title
Fixed date picker not opening, a regression from last week.
-
redesigned revision history, added 'checkpoints', diff view, and improved that ui a bit.
Some small bug fixes when it came to meeting recordings. Block menu correctly updates to allow 'summarize transcript' after a recording happens.
-
Improved tool calling and tool call 'memory loss' where it would have rerun past tool calls. More reliable edit_note tool calls.
Fixed several chat stability issues including the Apply button breaking chat on the home page and chat input disappearing after Apply/Diff operations. Also improved Diff All navigation with better auto-scroll behavior, fixed child blocks incorrectly appearing as top-level items on prompt pages, and simplified block tag handling.
-
Grug Speed model updated to Kimi K2.5 model on Fireworks.ai -- new SOTA open weights model
Bug fixes related to the block menu -- funny state issues when blocks were updated or lazy loaded.
-
Fireworks AI - Fastest Inference for Generative AI
Fetched from: https://fireworks.ai Date: 2026-01-28 07:37 Published: 2025-09-24
"Fireworks has been a fantastic partner in building AI dev tools at Sourcegraph. Their fast, reliable model inference lets us focus on fine-tuning, AI-powered code search, and deep code context, making Cody the best AI coding assistant. They are responsive and ship at an amazing pace."
"By partnering with Fireworks to fine-tune models, we reduced latency from about 2 seconds to 350 milliseconds, significantly improving performance and enabling us to launch AI features at scale. That improvement is a game changer for delivering reliable, enterprise-scale AI"
“Fireworks has been an amazing partner getting our Fast Apply and Copilot++ models running performantly. They exceeded other competitors we reviewed on performance. After testing their quantized model quality for our use cases, we have found minimal degradation. Fireworks helps implement task specific speed ups and new architectures, allowing us to achieve bleeding edge performance!”
"We've had a really great experience working with Fireworks to host open source models, including SDXL, Llama, and Mistral. After migrating one of our models, we noticed a 3x speedup in response time, which made our app feel much more responsive and boosted our engagement metrics."
"Fireworks has been a fantastic partner in building AI dev tools at Sourcegraph. Their fast, reliable model inference lets us focus on fine-tuning, AI-powered code search, and deep code context, making Cody the best AI coding assistant. They are responsive and ship at an amazing pace."
"By partnering with Fireworks to fine-tune models, we reduced latency from about 2 seconds to 350 milliseconds, significantly improving performance and enabling us to launch AI features at scale. That improvement is a game changer for delivering reliable, enterprise-scale AI"
“Fireworks has been an amazing partner getting our Fast Apply and Copilot++ models running performantly. They exceeded other competitors we reviewed on performance. After testing their quantized model quality for our use cases, we have found minimal degradation. Fireworks helps implement task specific speed ups and new architectures, allowing us to achieve bleeding edge performance!”
"We've had a really great experience working with Fireworks to host open source models, including SDXL, Llama, and Mistral. After migrating one of our models, we noticed a 3x speedup in response time, which made our app feel much more responsive and boosted our engagement metrics."
-
Added custom tasks in the block level ... menu. This allows creation of custom routine prompts that can be applied to any block. Pretty cool!
Added ability to "fold" blocks.
OH DANG. regression bug that cause duplicate word/phrases on some saves. If that happened to you in the last two days, sorry! fixed and wrote tests to avoid in the future.
-
more changes to meetings recordings and long/lock recordings. Now they are uploaded in progress, for better speed, and most importantly support for very long transcriptions. Still testing, but trying to make this increasingly reliable.
-
massive improvement on quick edit and ai/edit via speech feature. It's stateful now, so it has a bit of a short term memory. Successive edits work a lot better. Undo even works quite often. Also added a diff transition so the update isn't so abrupt. very nice!
-
Many many many tiny papercut level bug fixes deployed.
-
Some significant bug fixes. There were some hidden UI elements that were interfering with text positioning on mobile. Frustrating bug. Happy that's fixed.
Redesigned the block settings. And there's an MVP of ENHANCED NOTES, which is a little AI routine to take a long meeting transcript and convert it to concise notes.
-
ui improvements on mobile. trying to remove a lot of jankyness that's slowly crept in. needs work, but already better.
-
bug fix. block deletes are handled better -- bug was when a auto generated block was deleted, a page refresh was required for it to come back. now it just works.
bug fix. Add lightweight rendering for large prompts (150+ blocks). blocks older than a month. keeps page size reasonable for multiple years of notes.
-
big adjustment to audio ui -- can now switch target on the fly. also improvements in meeting and lock mode. and screen lock applied to upload phase as well.
-
pretty massive deploy of new audio features.
- The meeting and lock buttons now allow the rest of the app to be usuable while recording (don't click links however).
- The meeting uses Eleven Labs speech to text now, vastly better results for multi-speaker audio
- There is a new recording page under menu -- recordings are stored realtime locally, so much less chance of data loss! This is a big step!
- Meeting and Lock recordings now get put in sub-blocks automatically...
Will me making some adjustments over the next few days, but this is a big step!
-
added ability to rename a prompt! Added tabs and a little ui tune up on prompt settings modal.
Added a diff all button, ie so it's easy to inspect many block changes at once.
-
Some interface improvements in chat. Better model picker and small ui tweaks.
Some mobile ui fixes where keyboard would get in the way of the last block.
-
a little ui clean up of ai memory features.
-
Massive deploy. AI Memory added. Makes a huge difference in continuity over conversations. This is so cool.
Subscribe