ทวีตที่ปักหมุด
Wim Cools
1.1K posts
Wim Cools
@wcools
Building https://t.co/mH2S4ymRdI - A new IDE but for notes and tasks (local-first, encrypted and multiplayer)
Eindhoven เข้าร่วม Haziran 2009
286 กำลังติดตาม9K ผู้ติดตาม
Thanks! Do you mean memory for features in the app itself, or for the LLM? Haven't tried many local LLMs yet because my main machine doesn't have the best specs for that, so just using byok-AI requests getting forwarded to Claude. That then runs bash commands on the terminal/vfs in the app which doesn't really take extra memory. We keep the text for notes in memory but that's really not that much (I have hundreds of notes but those just contain a few MB worth of text), don't even think we need to lazy-load that. VFS is based on that data so very small, attachments we can load lazily.
English
Hmm so you have
1) the new version of thymercli right? (0.3.0-202602261145)
2) and you have thymercli open separately
3) while using the updated config for Claude Desktop, so without the 'http': { "mcpServers": { "thymer": { "command": "/path/to/thymercli", "args": ["stdio"] } } }
It could be a Surface Snapdragon issue but if the browser can connect then it sounds like it should work!
English
@wcools @NicolasDuch3min nope, can't get it to work after 45m. Browser connects fine by Claude Desktop just can't connect... Hmmm. Could there be an issue that thymercli.exe is running on a Surface Pro 11 - ARM/Snapdragon?
English
I was just wondering when you would release the MCP.
Even as a non-technical people, took me 5 min to configure with Claude Code.
Now, let's have some fun with #thymer
Wim Cools@wcools
Local AI agents really feel like a turning point. That's awesome but tbh I'm just as excited about the terminal/CLI aesthetic comeback they're causing 🤩
English
Wim Cools@wcools
Okay a very-high level (raw!) braindump of some architecture/stack decisions. We'd love to write some blog posts about it later but gotta ship first :) - app written completely from scratch in vanilla ES6 - no frameworks, no dependencies - (.. except libsodium for encryption and one for grapheme parsing if i'm not forgetting sth) - entire dev env is simply editing source file and reloading - CSS rules and design also done by hand - editor also completely from scratch. no contenteditable. - .. this was needed to seamlessly blend outlining (structured/networked) features with long form editing, so it both works without losing the illusion of simply editing text - that includes logic for scrolling, dragging, selections, (multiplayer) cursors, formatting, dataviews, ... - app UI plugins can be written in JS to create dynamic views or add features. isolated by subdomain. - we try to batch, defer and delta-render whenever possible. virtual scrolling in some places, lots more low hanging fruit there if needed. - main thread is for UI updates - custom-made sync engine, which runs in a separate shared worker thread - multiple tabs communicate async with shared worker (kind of offline multiplayer) - data cached locally in indexeddb ("sharded" into buckets for performance) - worker thread also takes care of other work like encryption. e.g. attachments are decrypted on first request off the main thread then cached in separate indexeddb. - because all data is local, things like dynamic queries, views and global fuzzy sort are pretty quick! - real-time updates over websocket, fallback over POST if needed - syncing backend server written in Go. - data updates synced as encrypted deltas, server occasionally creates new snapshots of entire encrypted state - both client and server consolidate deltas and resolve conflicts (like tree cycles) in same global order to ensure eventual consistency allowing for concurrent offline edits - server can be self-hosted on mac, windows or linux. deployment is running a single binary. - encrypted data on server is stored in a SQLite per workspace - encrypted files on server stored on FS or R2 - lowest part of the stack for cloud version: just simple metal, dedicated servers. commodity hardware super cheap and powerful these days There's plenty more for an actual write up but yeah those have to wait for now ;). If you're interested in more you can always ask @jdvhouten or me
QME
Turns out the "IDE but for notes and tasks" is also becoming my... IDE!
More and more I just make notes and boards with specs of what needs to be built, connect a bunch of code agents to thymer all working in parallel, have them pick up open tasks, build it and update progress back in the notes.
English
@garymc Great! Yep mobile is definitely something which is important to almost everyone. We're doing this in phases, first make it workable as PWA, then native apps later. But high on the importance list for sure ;). For your server, any chromium running there, or planning on cli only?
English
@wcools 1. Mobile - I have a shared Notion workspace with my wife. She uses her phone 99% of the time. Thymer's not ready on mobile yet.
2. An api/mcp for allowing an agent on a headless server to read/write. Looking forward to Thymer pub
Once these are in place I'll ditch Notion!
English
How frustrating, sorry about that, it seems Claude Desktop doesn't support MCP over http directly. I've just released a new version of the thymercli tool, along with instructions for Claude Desktop under "Setup" (#setup" target="_blank" rel="nofollow noopener">thymer.com/mcp#setup).
Basically, you start 'thymercli' (and leave it running), and then point Claude Desktop to use 'thymercli stdio'. The instructions describe what to add to your Claude Desktop config file (you might have to refresh the Thymer website, I've just added the instructions).
English
@NicolasDuch3min I got the #thymer cli running but Claude Desktop just couldn’t connect as MCP as either the port was in use / another cli thread blocking. Gave up after 45m… @wcools
English
It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December and basically work since - the models have significantly higher quality, long-term coherence and tenacity and they can power through large and long tasks, well past enough that it is extremely disruptive to the default programming workflow.
Just to give an example, over the weekend I was building a local video analysis dashboard for the cameras of my home so I wrote: “Here is the local IP and username/password of my DGX Spark. Log in, set up ssh keys, set up vLLM, download and bench Qwen3-VL, set up a server endpoint to inference videos, a basic web ui dashboard, test everything, set it up with systemd, record memory notes for yourself and write up a markdown report for me”. The agent went off for ~30 minutes, ran into multiple issues, researched solutions online, resolved them one by one, wrote the code, tested it, debugged it, set up the services, and came back with the report and it was just done. I didn’t touch anything. All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes.
As a result, programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You're spinning up AI agents, giving them tasks *in English* and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools, memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier "agentic engineering" feels very high right now.
It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others (e.g. especially for tasks that are well-specified and where you can verify/test functionality). The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. But imo, this is nowhere near "business as usual" time in software.
English
@tih4v @kylevamvouris Ooh this is really nice! Yes definitely want to make that easier. Already now there’s so many cool plugins I don’t even know of. What would you like to see most, easier in sense of installing and updating, a public gallery, both?
English
@kylevamvouris Thanks! Not for now, needs polishing.
I also hope it will be easier to share plugins when it enters public beta. 👀 @wcools
English
@wcools Looks amazing! Does Thymer support LaTeX for math equations yet?
English
@real_uncanny At the moment mostly Claude Code myself, and also connected it to Cursor. Check dm ;)
English
That's right, it runs in the terminal on the same device where your Thymer app (browser) is open. Next to "thymercli" we'd also like to build a "thymerlib", which would take care of things like syncing so you can use it to create headless servers running anywhere. But many other things to build first so that's a bit further down on the roadmap.
English
@wcools Looking forward to building some plugins using this. By the way am I correct in saying that this needs to run alongside a GUI browser? Or is there a way to have this run on a headless server?
English