Luan Nguyen

@kcosr Wow, how is it possible? How did you set it up?

More fun with Codex. 🤣

Codex CLI 1.116.0 supports connecting the TUI to an app server. You can even connect multiple TUIs and the desktop app at the same time. Set [features]:tui_app_server = true to enable.

@jakemor I've just asked codex to scan through the code, seems like it's possible with some change? Basically can parse claude/codex history for session id, then Kanna can send a resume request? Are you open for PR btw? Thanks for an awesome project.

@peeomid No but you could literally copy the convo and send as a message it’ll do just fine. Or ask Claude to point you to the transcript

Introducing 🌸 Kanna – a web ui for Claude Code + Codex running right in your browser (and my first open source project!) It's sort of like if Claude Code & Codex had a baby with a much better ui/ux 💅 Here's what makes Kanna special: 🔀 One-click switch between Claude Code / Codex 🌐 Runs in your browser on localhost - no app switching 🧩 Embedded split terminals, persisted between chats with a beautiful horizontal scrolling UX 📁 Full project file browser built in, embedded editor coming soon too. Coming soon: 🔒 End-to-end encryption + remote access via reverse proxy 🌿 Git integration, diffs, PRs 🧱 Plugins I wanted to replace switching between cursor, the github app, browser & (most recently) the codex app. The browser is hardest to recreate yet also the most flexible so it makes sense to build in the browser itself. Install in 5 seconds: bun install -g kanna-code then just type: kanna Kanna is purely a ui/ux layer. It uses your existing CLIs, is 100% compliant & no data ever leaves your machine. If running claude or codex in your terminal works, kanna works too. Kanna natively supports every tool call, model, reasoning effort, fast mode, plan mode, compaction, user questions, web searching, skills, agents, mcps, everything. There are a few libraries that do this but none that I found as comprehensive. All feedback welcome!

@Mho_23 How do we know if that’s not an actual human video?

here's another AI UGC video from our new system our new system is extremely good at details: > handles accurate product placement > realistic voice > stable/controllable movements > infinite length can make them at scale & FAST if you know what you're doing best time to be alive ngl..

@fayazara Awesome, thanks Fayaz

@peeomid I think I missed it. Working on the auto update from the app, should be out in a couple of hours.

Kaze v0.5 is out github.com/fayazara/Kaze/… There's a lot of new features in this release. You can finally pick microphones. You can see this new notch that you see on top and a major performance improvement. If the app is idle, then the AI model will be offloaded from your memory.

@elvissun @openclaw Ah that makes it a lot easier, thanks

@peeomid @openclaw credits are pooled

best coding stack as of Feb 2026: - agent swarm via @openclaw, self-monitor via cron - backend/frontend: gpt 5.3 codex xhigh - UI design: gemini 3 pro - managing git: opus 4.5 - 3x team plan with ChatGPT ($90) - Claude max ($100) - Gemini via cli (free) x.com/elvissun/statu…

@ValkDokk @pradeep24 What do you use to check this?

@pradeep24 Yeah about that

@christou_c @jdrhyne @mvanhorn I have a folder with skills in there, then sim link that folder for all agents’ skill folder, so only need to add new skill to that, will be available for all

@jdrhyne @mvanhorn Do you manually add/update then across the three agents?

Update: 20 skills → 32 and counting. Added this week: • Salesforce, Gong, Zendesk — CRM & support stack • knowledge-graph — Three-Layer Memory System for persistent agent context • last30days — research what the internet is saying (tweaked off of @mvanhorn skill) • parallel-task — run subagents simultaneously (credit: @LLMJunky) • frontend-design & web-design-guidelines prompts 77% work across @openclaw , Claude Code, and Codex. Contributions welcome 👇 github.com/jdrhyne/agent-…

@AntoineRSX How do you set it up for web search and get result from that?

Here's a quick rundown. Everyday: • Web search & fetch pages • Set reminders & scheduled tasks • Weather forecasts Productivity: • Manage Apple Notes, Reminders, Things 3 • Google Workspace (Gmail, Calendar, Drive) • Notion pages & databases • GitHub issues, PRs, CI Communication: • Send iMessages, WhatsApp, emails • Post to X/Twitter • Control Slack Creative: • Generate & edit images (Gemini) • Text-to-speech (ElevenLabs) • Transcribe audio/video • Edit PDFs Technical: • Run shell commands on your Mac mini • Browse the web & automate browsers • Spawn sub-agents for complex tasks • Control MCP servers

@iamtrebuh @openclaw For 2, not exactly the same, but I have multiple discord channels for different topics, each has different system prompt, then I can talk to them all at once

I have two main concerns about @openclaw. 1. When I need the best output possible from my model, for example for some business related brainstorming, wouldn't clawdbot's memory system fill the context with too much unrelated stuff that could decrease the quality of an output? 2. When I'm using Claude Code, I can easily run multiple agents, I see all of their output at the same time and it's easy to talk to all of them at once, and if needed, it's easy to close them or open a new ones. With clawdbot, I haven't found a solution for that. Is there any? Or are we a little bit more limited to the amount of working in parallel agents? @steipete

@steipete Is it possible for this to show active sessions’ status? I normally have multiple sessions of codex/claude code running in parallel, so have to switch back and forth between iterm tabs just to check if it’s finished.

Codexbar 0.7.1 is out - now with handy shortcut for switching accounts. github.com/steipete/Codex…

@deeflectcom @omarsar0 Thanks, will check out your blog

@peeomid @omarsar0 I did some amount just rags on my blog, but not particularly about self learning agents. But simplified workflow is you just use RAG for “memories”, and can chain it with other cheaper model to evaluate decisions and seed the memories

Fine-tuning LLM Agents without Fine-tuning LLMs Catchy title and very cool memory technique to improve deep research agents. Great for continuous, real-time learning without gradient updates. Here are my notes:

@phuctm97 @cursor_ai Give windsurf autocomplete a try, it’s better, and free

Just re-subscribed to @cursor_ai Pro plan, their AI autocomplete is just unmatched.

@deeflectcom @omarsar0 Do you happen to write about it somewhere or have an example i can have a look? Thanks

memory-based agent improvement is where the real work happens most people are still stuck on prompt engineering when the actual breakthrough is in how agents learn from their own execution patterns. been building this approach for months - way more reliable than constant fine-tuning

@dayonefoundry Have you tried run init to generate claude.md? will help it understand project. Should add your common practice there as well. Then what normally work for me for bigger tasks, is to ask it to plan it, fine tune plan, then write that to a file, then implement.

@peeomid just talking to it like i would with cursor. I thought i could give it larger tasks or a list of tasks. But found it's better to micromanage small tasks one at a time. Which is how I work with cursor.

Finally gave into all the hype and starting using Claude Code and... I have 0 productivity gains over using Cursor. I actually think I have more frustration than before. Not to mentioning getting cut off after 30 min. What am I doing wrong?

@shaoruu @cursor_ai Applying is a lot better now, but sometimes it still fails, would be great to have it more reliable.

focusing on these areas for @cursor_ai composer: - easier and faster way to get relevant files - reviewing all changes across multiple files - making composer more powerful (e.g. fix its own errors automatically) - make it clearer when to use chat v.s. composer

@JulianGoldieSEO Yes