Brandon Z. Hoff
Region: US
Wednesday 16 July 2025 20:22:36 GMT
25185
711
101
41
Music
Download
Comments
jeekbeen :
None of you use open models?
2025-07-17 04:11:32
1
Mr. Aitch :
Have Claude [echo ‘{ … detailed context … } > context_dump.json] without a summary. You can even setup an alias for this and other agents can load it. Set up templates for the context structure. The process takes a few seconds.
2025-07-16 22:36:11
46
FAF0 BACON :
I saw somewhere where someone uses Gemini with Claude Code to use the Gemini 1M context window but Claude in CLI. Haven’t figured that out yet, if you do - please tag me :)
2025-07-18 20:31:06
1
Bujavi :
2.5 hours on the $200 plan? I'm getting like 4-4.5 hours on the $20 plan. How are you using it? 😅
2025-07-17 07:34:37
4
Slothsbeslothing :
I’m on the broke $20 plan and noticed that from day one and have been doing exactly what you are saying but using a project scope-progress.md and I have that referenced in the Claude.md file for operating guidelines.
2025-07-16 20:53:07
1
jordanlister :
Opus or sonnet?
2025-07-23 09:57:50
2
Jonathan Edwards :
-give Claude desktop file system access, use it to anylize and plan then tell it to instruct Claude code in a shared.md file.
2025-07-16 21:57:50
1
Justin Michael :
claude was getting throttled massively today
2025-07-17 03:16:14
13
The AI Syndicate :
If you have time, go look at the video i just uploaded. i built a mcp memory, and it completely removes this issue. the video is showing claude desktop, but my claude code is hooked up to the same MCP server so they also share that memory. i work in claude desktop until i hit my limit then switch over to claude code and run the API per token. No downtime, no updating context in new threads or projects.
2025-07-21 03:49:10
1
Drew Parker :
$200 usage plan and u hit ur max limit??? I’m on $100 plan and code all day and barely hit the limit, (non-opus) how big is your codebase?
2025-07-16 21:46:13
4
John McLaughlin :
In the last 2 weeks it has for surrrrrre started throttling, I used to never hit the usage limits on $200 plan and now it hits it absurdly fast. And 100% the context window gets siphoned by auto-compact, summarizations, and wasted md files. Completely ignores Claude.md root instructions now too
2025-07-16 22:37:56
5
theaiguyatl :
I had to just end up paying for straight api usage
2025-07-16 20:53:42
2
Davo :
You can build a matrix table of the token limits you want Claude Code to use, on a CLAUDE.md. Also, you can make an observer agent that taps into the hooks, and halt Claude when it overthinks. Overthinking is a signal that you need to break the context into smaller spaces.
2025-07-16 23:49:47
1
Christine23 :
You’re always 10 miles ahead of me. I haven’t migrated to Claude. Still using ChatGPT
2025-07-17 00:03:10
1
Robert M. 🇯🇲🇺🇸 :
Mine will still auto compact. I’m on the $100/mo
2025-07-16 22:51:59
2
Spiderman824 :
I make a ton of .md files, and have the Claude.md have instructions to continually update them. Things like current_session.md that I tell it to continually update. Plus larger project plan and todo lists in Md’s I actually make it break things into sprints and add the sprints to a main Md then ask it to break off sprint specific Md files while we’re working them. So, yes, I do this, to the extreme
2025-07-17 04:56:37
3
Holden Russell :
Yes the markdown checklist is super helpful. I tell it to make a plan in .md format before executing then update the plan as it works through it
2025-07-17 02:25:28
4
bluesdave11 :
So when I’m in ChatGPT, when the context window is too big I converted to a file. Then I start a new chat, and I say go read that file, and bam it speeds up, and has access to all the old conversation.
2025-07-17 03:36:38
2
Bert Rico :
I use Cursor, haven’t dedicated the time to test out CC yet, but I typically make the debug report and todo list as part of my project rules in the MDC file. I noticed that not only helps be more token efficient, but also it produces better quality code so less for me to QA later.
2025-07-16 22:53:42
1
Beady Eyes :
It is over hyped I used and nothing special
2025-07-17 20:13:36
1
brocka11 :
Omg!! I’m on the 20 dollar plan. This is so helpful.
2025-07-16 20:43:37
1
Snackk.sol :
I’m on 5x plan and can code in one window for at least 3 hours on each reset
2025-08-17 13:02:10
0
AIorBye :
I have not run out at all
2025-08-18 11:12:56
0
Bryan Batchelder :
Keep in mind that every time the LLM is called, you’re sending the entire message history. So the longer a session is, the more tokens you burn on each request. So you proactively pruning it is keeping you using less tokens overall, so you don’t hit your token limit as quickly.
2025-07-23 05:06:03
0
Shavoni :
Debugging Reference Prompt: When I need to modify or debug elements, I will reference them by their unique IDs. Always maintain this ID structure throughout development and updates. If I say "modify the #hero-section" or "change the styling for #user-profile-card", immediately understand I'm referring to the element with that specific ID. When making changes, preserve all existing IDs unless explicitly told to rename them.
2025-07-27 03:39:04
0
To see more videos from user @hoffdigital, please go to the Tikwm
homepage.
