MostLime

Shipped v0.7.0

Changelog:

Changed

• Removed default 4-iteration limit for tool calls (now unlimited). Set tool_max_iterations in settings to restore previous behavior
• Rewrote streaming pipeline with requestAnimationFrame batching for ~60fps rendering and reduced latency
• Enhanced health check endpoint to return quota/balance details for better error messaging
• Extracted duplicate message serialization logic in conversation store

Added

• 60-second stream timeout to detect stalled connections

Fixed

• Fixed undeclared hasPDFAttachments variable causing potential reference errors
• Fixed tool argument parsing to return errors to model instead of executing with empty args
• Fixed API health check to properly display quota exhaustion messages

I’ve shipped versions v0.6.1 and v0.6.2, both patches. Among the two, I’ve made many small back-end improvements like bot detection (on vercel) and error handling improvement & api health checks, but I’ve done plenty more too.

Full changelog:

[0.6.2] - 2026-01-27

Fixed

• Improved error handling and added API health endpoint to better catch when the API is down
• Improved caching and model context handling by moving the current time from user prompt to system prompt
• Added models (Kimi K2.5, GLM 4.7 Flash)
• Replaced Gemini 2.5 Flash to GLM 4.7 Flash for conversation title generation
• Replaced default LLM from Kimi K2 to Kimi K2.5
• Improved memory tool reliability
• Polished system prompt to better instruct model to use memory tools

[0.6.1] - 2026-01-25

Fixed

• Added models (Grok 4.1 Fast, MiniMax M2.1, GPT-5.2, Gemini 3 Pro Preview)

I just fixed the app cuz it wasn’t working before. This time I added limits (48 messages a day & 8 image generations a day) to avoid people spamming stuff, though it’s a shitty implementation cuz I want to redo everything later. I also added chat branching.

Full changelog:

[0.6.0] - 2026-01-23

Added

• Added limits (48 message requests per day, 8 image generations per day)
• Added ability to add your own HackAI API key with no limits
• Added message branching with editing/regenerating messages

Fixed

• Kimi K2 non-reasoning now functions as intended and does not route to Kimi K2 reasoning
• Errors will now be visible to the user for enhanced debugging
• Removed deprecated models (Gemini 3 Pro Preview, Gemini 3 Pro Image Preview)
• Enhanced visual polish

v0.5.0

Added

• Added models (Gemini 3 Pro Image Preview, Gemini 2.5 Flash Image, GLM 4.7)
• Added full image generation functionality
• Reasoning and tool use is incorporated into model context

Fixed

• Refactored the part system, resulting in more reliable streaming & tool use and fewer gaps between tools in the UI
• Annotations from documents are now properly stored and passed into model context
• Images are stored in model context throughout the chat without need to reupload them again

Migrated from github pages to cloudflare pages, which required changing up the plugin configuration n stuff. I also experimented with smarter optimization.

Also, the website url is way shorter now, no more mostlime12195.github.io, but just mostlime.pages.dev.

Heavily optimized performance, while improving structure.
Performance improvements:

• Removed/moved unnecessary styles
• Inlined & minimized the CSS so the browser doesn’t have to make several calls to download the CSS (the CSS is built into the HTML, before it would have to download the HTML and afterwards the CSS, which added a delay. Unfortunately now there is no caching, but the stylesheet was small enough not to need it.)
• Preloaded the Merriweather font

Structure improvements:

• Fixed elements that had no closing tag
• Formatted the code
• Added description metadata
• Modified header structure in order to be more accessible to screen readers

HOWEVER, we’re not 100% optimized yet! There’s one more goal left, and that is to remove unused CSS from pages that don’t use it, aka tree shaking.

I’m not going to implement that manually, of course, because I’m lazy. I would’ve done it with a Jekyll plugin but whoops, GitHub Pages doesn’t support the plugin I need, so that’ll be for another time.

The attached image is the pagespeed.web.dev analysis of my website.

*fyi about 2 of these hours were for the previous ship, but I had forgotten to log them, so I’m logging them now.

I’ve polished up the blog quite a bit so far. I’m now working on the individual post pages. I’ve mostly completed the styling, and am polishing it.

The website is still 100% JS-free, so hurray!

I’m creating a blog.
And better yet.. the project will remain javascript-free!

I’m using Jekyll, which will compile all my markdown & dynamically create the HTML & paths that’ll be used to showcase my blog posts easily and efficiently, WITHOUT javascript. All this means is that the site will be lightning-fast and will load super quickly.

I finished up the first version of the personal site, and shipped it!
I did a quick revision of the text content, added a bit more, and added it to github.

Next up, I’m planning on making a blog for this!

My personal site is almost complete!
I’ve made major modifications to the font (now Merriweather instead of Georgia) and leaned in a bit more onto the newspaper style, with a cool masthead.
I made a couple other subtle polishes and changes. The site will likely be shipped by my next devlog.

Added a bit of content!
I’m thinking I’ll make a small (and straightforward) personal site just to avoid me losing interest (like every other time :[ )
I’m liking the theme, it just feels too simple.

Begun work on my personal site! I’ve always wanted to build one, but I’ve never had the motivation to do actually make one. I decided to take a break from AI and make this project with my own two hands (just like old times!)