Pavlo

@shipilev are you sure it is not mythos already? :)

I honestly would enjoy either outcome, whether it runs into the ground, or takes off brilliantly. But I suspect the real outcome would be limping through the lingering issues for quite a while. The timing is just right (chef's kiss) for Mythos to come out of closed preview.

Ship it! (I have no words.)

@WebReflection Bun is not the only project written in zig :)

I wonder if Zig will fade away at this point 'cause Bun migrating from it means Zig is not as cool or mature people want to believe it is ... inferior debuggability is *huge* specially in these AI (slop-driven-development) days ... crashes with no meaning are very hard to fix.

@mitsuhiko @maddada @robinebers how is it different to the loop extension that you had?..

@maddada @robinebers I pointed pi to codex and asked it to build a version of it. I did not use it, because I did not use /goal in codex yet so I first want to see how it works there. github.com/mitsuhiko/agen…

Let's see how this /goal thing in codex' harness is doing.

@tanishqk I wonder why this instead of github.com/nicobailon/pi-…?

I built pi-live-terminal - a live view of any persistent or long-running commands, powered by Just™ tmux > you or the agent can start and manage panes > shortcuts to focus, kill or close panes > LLMs already know tmux and can interact with panes natively Here's Pi running inside of itself. cc @badlogicgames Use it for: - dev servers - interactive commands like publishing to npm - - running tests in the background - Having Pi test its extensions, visible to you

@badlogicgames What's the reason/end goal for this one?

People of pi.dev. I'm heads down in the big refactor (see bigrefactor branch). To have any chance of finishing it, I must pause issue triage for the next 2 weeks. All issues filed during that period will not be reviewed. In case of emergency -> Discord

@mitsuhiko npmjs.com/package/ws claims it does support per message-deflate. are there strict reasons to avoid it as a dependency?

@fxposter Codex uses permessage-deflate for messages to the server. The websocket library in node doesn’t support that. It also forces max size of 4MB post decompression.

I have a PR up to fall back to SSE on WebSocket errors in pi. If someone runs into these, please give this PR a try and report: github.com/badlogic/pi-mo…

@FredKSchott @tanishqk @threepointone I wonder why pi is not mentioned anywhere (unlike CC and codex) except in the actual code of the sdk? :)

@tanishqk @threepointone Framework vs. SDK. Kind of like how Next.js/Astro is built on top of React, Flue is built on top of pi-agent-core to power our harness. Then we offer a different, more holistic DX on top of it. x.com/FredKSchott/st…

Introducing Flue — The First Agent Harness Framework Flue is a TypeScript framework for building the next generation of agents, designed around a built-in agent harness. Flue is like Claude Code, but 100% headless and programmable. There's no baked in assumption like requiring a human operator to function. No TUI. No GUI. Just TypeScript. But using Flue feels like using Claude Code. The agents you build act autonomously to solve problems and complete tasks. They require very little code to run. Most of the "logic" lives in Markdown: skills and context and AGENTS.md. Flue is like Astro or Next.js for agents (not surprising, given my background 🙃). It's not another AI SDK. It's a proper runtime-agnostic framework. Write once, build, and deploy your agents anywhere (Node.js, Cloudflare, GitHub Actions, GitLab CI/CD, etc). We originally built Flue to power AI workflows inside of the Astro GitHub repo. But then @_bgiori got his hands on it, and we realized that every agent needs a framework like Flue, not just us. Check it out! It's early, but I'm curious to hear what people think. Are agents ready for their library -> framework moment?

@jeremyphoward github.com/nicobailon/pi-… with gemini. tried perplexity there, didn't like it much. also, plan to try out @mitsuhiko's github.com/mitsuhiko/agen….

What search providers are you all using with openclaw/pi/opencode/etc? Brave; serpapi; gemini; ...? Got any favorites?

@samsaffron they mention it in openai.com/index/introduc…: "In Codex, GPT‑5.5 is available for Plus, Pro, Business, Enterprise, Edu, and Go plans with a 400K context window"

TIL ... unlike 5.4 the 5.5 endpoint on codex is limited to 272k context. Which is fine, just something I need to update

@theo I actually think that 5.5 would not have "cheaper short context" at all and all requests up to 1m would be priced the same. we'll see when they update pricing page. but yeah, for most existing flows it would be 2x more expensive per token.

$5 per mil in, $30 per mil out. GPT-5.5 is smart. I've been using it for a bit. It's also weird, hard to wrangle, and too expensive IMO. Double the price of GPT-5.4. 20% more expensive than Opus 4.7.

@mitchellh ah, so you're not using it via harness LSP integration, but instead querying "directly" by the LLM. didn't think about it. though, "go doc" is what gpt models know how to work with, too.

@fxposter Saving a ton of tokens for finding things like interface implementations, API callers, etc. You don't even need an agent skill, a one-paragrapher in your AGENTS.md thats basically like "for finding APIs or callers use gopls" works.

I'm writing Go again (for what, you'll see later...). `go doc` and `gopls` are like agent superpowers and its shocking how productive agents are out of the box at writing [good] Go code versus other languages I've used (including the JS ecosystem). Also, Go + Zig is a good mix. Go for the higher level and concurrent stuff and then no-libc Zig code plus the Zig compiler for zero dependency cross-compiled cgo with high-performance characteristics (minimize cgo boundary crosses). Chefs kiss. Its funny because a lot of the shitty ergonomics of Go CLIs like `go doc` and `gopls` (prev. stuff like `go oracle` or `guru`) are totally obviated by agents and not just that but in a twist of irony they're excellent for agents. Don't worry, its not Ghostty. Ghostty and libghostty will remain pure Zig; it's a fantastic fit and a perfect pairing. This is for something else. "Wait, I thought you said Go has no place anymore?" I was wrong, mostly because agents are so productive at Go. I won't bring in other languages in this discussion because I don't want to feed the crabs, so to speak. lol.

@Zenoware @ImSh4yy and what does it mean? models.dev/?search=kimi+k… context and output limits for opencode and cf are different, btw. also, I am not sure what it prooves.

@fxposter @ImSh4yy models.dev is where opencode fetches its model list

Wait, Cloudflare is now offering Kimi K2.6 at the same price as Moonshot AI but with much lower latency and higher throughput?

@Zenoware @ImSh4yy how do you know that?

@ImSh4yy They're also the current provider right now that works out of the box with OpenCode.

@mitsuhiko What you ended up using?

There is VoiceInk and there is Handy. Is there more? What's your favorite Open Source dictation app and why?

“Stop” might be the right word for concurrent collectors that aren’t on-the-fly, and even then it’s kind of sketchy. Folks often end up with concurrent GCs where the time the mutator is “stopped” is less than the time that a C or Rust mutator would be “stopped” waiting for mmap to return the next available page for an allocation that is growing the heap. For example in FUGC (which is on the fly), the closest thing to “stopping” is a thread scanning its own stack. But what if C++ exceptions have to scan the stack? We don’t say C++ “stops” for that, even though C++ exception unwinding is slower than FUGC stack scans The discussion of “stopping” and “pausing” is detrimental to a good understanding of GC overheads and *why* GCs are often a bad idea. The reason why they’re a bad idea: - GCs will use more memory than a program that uses malloc/free. - GCs cause memory leaks. Relatedly, GCs make it impractical to estimate max memory footprint in systems where this needs to be known statically - GCs increase page demand (the rate at which a program asks for a page that isn’t mapped). - GCs cause a tremendous amount of memory traffic. (A concurrent on the fly GC will have an effect on latency that looks like a pause because during the time it’s running, it’s slamming the memory subsystem). - GCs have positive feedback. They are likely to be triggered when the system is under stress. Hence, they make stress conditions worse. - GCs tend to be inelastic. That is, programs that suddenly free up a bunch of memory will tend to sit on it because the GC hadn’t run yet. - GCs play poorly with real time locking and scheduling (even real time GCs that have a story for this have a bad story in the sense that schedulability analysis gets way nastier than it would be otherwise) These are all good reasons to avoid GCs. Folks should know these reasons. Folks working on GCs should try to address these reasons directly. Talking about “stopping” and “pausing” just confuses the situation by making it harder to see the real problems.

@thsottiaux Is it only for codex app? If so - any plans on making it available in cli?

Codex Compute efficient ✅ Always up, never down ✅ Best at hardcore engineering ✅ Crazy good app, first to escape the terminal ✅

@mitsuhiko what's the difference between this and github.com/mitsuhiko/agen…?

We moved our /review extension for pi that we use at Earendil into a public repo. github.com/earendil-works…

@0xSero I feel that this will finally be resolved only when fixed-price subscription are over :)

@fxposter Why is it a gray area? It’s been a year. Wouldn’t it be professional to give out a clear concise description of what what usage is and isn’t allowed?

I figured out how to get Claude working anywhere without extra usage and technically in line with Anthropic's TOS. I don't recommend you adopt this, might get you banned but as you can see, no errors, no extra usage consumed. The point of this is to demonstrate futility.