rcarmo

The Science Fiction section is pretty much spot on, and in roughly the order I’d do it myself :)

Why the downvotes? Are people completely oblivious to https://github.com/huggingface/tau ? That far removed from the fact that the harness is not the TUI?

Tau is kind of taken as a name, BTW, by the Python harness from Huggingface. And I do agree Pi is great, but what makes it great is the back-end (pi-durable, pi-ai, etc.).

I've been working on https://github.com/rcarmo/piclaw for most of the year, as well as https://github.com/rcarmo/gi and https://github.com/rcarmo/go-ai (as ports of pi and pi-ai/durable/etc., plus 3 or 4 pi-durable add-ons for Obsidian and other desktop apps) and the internals of pi are a grand thing indeed, because it was designed from the outset to be able to extend itself (which many people miss, because many people confuse the harness with the TUI).

Damn. And me here typing "make it so" to AI agents instead of re-watching TNG.

Let me go and fix that.

Also, one of the most... colorful reviews. I watch his stuff just because of his style...

I was just looking at https://claudia.gallery and the MCP specs there (which reminded me instantly of Idoru) when I chanced upon this, and the Gibson quotes.

Yep. He _has_ to come from the future.

Love it. I too use Platinum as a skin in my projects: https://github.com/rcarmo/wrdp

This is brilliant. Just brilliant.

As someone who works in the field… having models that do something is not the same as having a workable solution.

I think we’re going to see a lot of this clueless cloning across the board. Photoshop and the like are very visible targets because they’re popular and the “outcomes” of this cloning are flashy and trigger endorphin responses in people who think “it’s just about coloring pixels, how hard can it be?” But as the article says the core engine is flabbergastingly efficient - I ran Photoshop on early Macs, and boy, was it devastatingly efficient for the hardware at the time.

That is nowhere on these projects’ roadmap, which is kind of sad because AI is pretty decent at targeted optimizations, if you know what to ask for and how to guide it.

Many people are using Nemotron (NVIDIA provides Linux binaries for ARM and Intel that you can shove into a server and do audio over websockets for them).

As much as I like seeing this, the space is really crowded, and I think the real value is not in single dictation for input, but in two other things: diarization (essential for call recordings, and a staple in video calling services, but hardly touched for in-room meetings and brainstorming) and the next step, which is cleaning it all up for actually useful notes. Not just the transcript, but better variations on meeting notes that you can customize, link to previous existing information, etc.

Yeah, it's so good I was seriously considering stopping maintaining piclaw and moving to it. But the... contribution policy isn't there (one PRs and Twitter discussions didn't go anywhere) and my dream of maintaining _less_ stuff didn't pan out... I don't want to fork it and fall down the rabbit hole of still maintaining something, so I will just slow down piclaw maintenance to monthly releases (minus hot fixes and bugs, of course) and see how that goes.

At least that way I can automate deployments and have extensions for each agent (and working networking, as well)

Feelis like working at Magrathea...

Well, as long as it doesn't start developing anatomically accurate metal skeletons with red glowing eyes...

Have a look at https://github.com/rcarmo/gi - I am slowly cleaning it up to upgrade to the equivalent of pi 1.0.4

I keep wanting to use LISP, which is kind of why https://github.com/rcarmo/gi uses https://github.com/rcarmo/go-joker, my IR-accelerated Clojure interpreter (which can also do WASM, and thus isolates).

I end up using Go for mostly everything, though, because a) Python is too slow and b) Rust will regularly fill up the hard disk of _every single one of my sandboxes_ with just... too much stuff, and takes around four times as much to compile (I can build, link, profile and fuzz Go in the same time I get a plain Rust build).

Still, I hope that one day I will be able to do just LISP :)

Marketing is tough. I have a pi-adjacent integrated environment (https://rcarmo.github.io/projects/piclaw) and it might as well not exist in that ecosystem :)

I built https://rcarmo.github.io/projects/piclaw atop it. It's pi, but with a web UI I use every day, across dozens of agents, and has pi-based add-ons. I do use the TUI on occasion, but need a remote web UI for 99% of my stuff.

I'm loving it. Already converted a few of my smaller tools to it, and am ripping out the guts of piclaw to replace them (in time)

Well, check out rcarmo/gi - I am using my clojure interpreter inside it as a scripting engine.

Well, I'm happy. Thanks for the new harness, ripping out the guts of piclaw to use it, and also flipped a few other small tools to pi-durable, which is nicely streamlined.

I can't for the life of me figure out why people would think pi is bloated.

Neat, but I keep having to edit the pi stub to remove the "/bin/env node" and replace that with bun instead, because, well, somehow that's still hardcoded.

The official one seems to do, as well as other info from the engine (I think remember their mentioning LOD/UV map hints in one of the public demos, or articles, a few months back--or it might have been an Unreal Engine podcast)

Addendum: My unfettered notes on chaining MCP operations are here: https://github.com/rcarmo/umcp/blob/main/docs/CHAINING.md

I happen to think codemode is useful, but not the full answer. I have a long and skewed history with chaining things in MCP and built a dozen or so enterprise ones (see https://taoofmac.com/space/blog/2026/04/29/2341 for notes) and it all falls back into the trade-off between agent scope/context and tool coverage: If you are using a coding agent it will have no trouble sorting out any tool regardless of how many are exposed (it's just a matter of either progressive tool disclosure or good tool metadata, since the coding agent will just go at it and expend whatever tokens are needed), whereas in a "normal", limited, scoped agent that has only a few things it needs to do (like handling a ticketing system) codemode is pretty much overkill.

Pi is primarily a coding agent, so yeah, code mode makes sense, but I've found that better MCP design saves everyone a lot of trouble and would also probably have improved the thing's reputation overall (I personally am not fond of the line protocol, would rather have protobuf and more typing, but it is what it is).

In enterprise integrations, that is just not an option. MCP has pretty much taken over there.

Have a go at https://github.com/rcarmo/memento, I would appreciate Claude testers since I mostly use Codex. Just trying it and filing an issue about what doesn't work would be great...

You can turn those tools off. In fact, that is what I am doing right now in https://github.com/rcarmo/piclaw until I am positive the new MCP stuff has full parity with the MCP adapter I've been shipping for the past six months or so.

I'm actually pretty happy that they did it, since 90% of what I have to integrate in enterprises is MCP-driven (it's a security and auth boundary that has become pretty much mandatory for any third-party agents wanting to reach into corporate data) and this lets me use Pi directly. Am just being cautious about the first version, because, well... it's a first version, and I like my tools stable.

(I actually played around with the idea of using QuickJS myself for codemode, but since I rely on Bun that gives me the ability to use other things... never got around to do it though.)

I have had to hack a couple of workarounds in https://github.com/rcarmo/memento to do uploads, and there's a draft going around, but the general practice in enterprise MCPs seems to be to do it "out of band" and have MCP tools to hand-over storage handles/URLs so the MCP server can do the imports itself "safely".