a tiny schemer. engineering @trydaily and @pipecat_ai core maintainer.
Greater Los Angeles Area
Joined November 2009
- Tweets1.8K
- Following67
- Followers519
- Likes1.9K
A couple of weeks ago, we released a demo with Pipecat and Unreal Engine: youtu.be/xfyRDiv5eWQ
Some people asked, "What about Godot?"
Well, it's Godot time! Happy haunting! 👻
github.com/pipecat-ai/pipeca…
@pipecat_ai @godotengine @UnrealEngine
Aleix Conchillo Flaqué retweeted
This is awesome. Multi-character interactions in Unreal Engine, completely LLM-driven and unscripted. You can spin your own version of this up pretty easily.
This is another Pipecat + Jev experiment from @JonPTaylor.
Watch Jon walk around the game world, talk to the characters, and ask them to do things. They interact with Jon, with each other, and with the world.
Tech stack:
- One Pipecat agent process syncing state with Unreal via the Pipecat RTVI protocol.
- Three non-player characters. Each character is a Pipecat worker with its own LLM prompt and tools. Each character also has its own TTS service so its voice is unique.
- When Jon (the human) talks, Jev assesses Jon's intent and which character(s) he is talking to.
- Unreal 5.8 with Spatial Audio. (Watch the video with headphones on!)
- Pipecat C++ client embedded in Unreal Engine.
Probably the best battleship game ever made. 🔥 🚢
NVIDIA released a fast speaker-labeling model today: Nemotron 3 Diarization. Like all of the Nemotron family from @NVIDIA_AI, this is an open weights model, so you can grab the weights from @huggingface and use the model in the cloud or on your local device.
Nemotron 3 Diarization is a useful building block for voice agents and realtime AI applications that need to handle speech from multiple people at the same time. It handles overlapping speech well, can identify up to eight speakers, and pairs with any transcription model you're using in your multi-modal pipelines.
@jonptaylor built a demo (code below) and recorded a technical deep-dive showing how to use this diarization model together with a speech-to-text model.
Jon's code also uses Jev for super-accurate confirmation of critical voice input in a noisy/multi-speaker environment.
Aleix Conchillo Flaqué retweeted
Audition mode is coming along in Jamcat. The generated midi guide is used in audio-to-audio to generate stems that match your voice prompt. Each instrument type needs tuning and training, but it's not far off!
Aleix Conchillo Flaqué retweeted
Here's a complete voice agent tutorial focusing on a super low latency configuration: Pipecat PhoneLLM Alpha 1 on @modal, Deepgram transcription, Cartesia voice.
You can grab the code and jus run /setup in Claude or Codex.
But Merve's video also dives into the details of configuring model endpoints on Modal, the Pipecat bot file, and the front-end UI code in the example project.
Aleix Conchillo Flaqué retweeted
Introducing PhoneLLM, an open model for voice agents.
GPT 5.6 Terra performance on typical voice agent tasks at 1/3 the latency and 1/18 the cost.
For voice agents, we need models that are both very low latency and very good at tool calling and instruction following.
There's a trade-off here, and we often have to compromise on either latency or capability when building voice agents. With PhoneLLM (and the training and data stack that made this model possible) we're fixing this problem.
For the last couple of years, most of the effort in frontier model development has gone towards leveraging test-time compute. Which is awesome! Models of all shapes and sizes are available that perform really, really well ... if you have "thinking" turned on for your model.
But if you need your agent to respond at voice conversation speed, you can't use thinking models.
PhoneLLM is a full-weights fine-tune of NVIDIA Nemotron Nano 30B. We trained on a wide range of real-world telephone and customer support use cases. The training focused on taking the excellent Nano 30B base capabilities and teaching the model to do typical voice agent tasks with thinking disabled.
The results are really good: accurate tool calling and concise, on-topic responses in long conversations.
And fast: TTFAT measured server-side is <100ms if you run PhoneLLM on a lightly loaded B200. :-)
But seriously, when we characterize model latency, we do it with full, end-to-end, batched request simulations using real Pipecat voice agent pipelines.
You can serve more than 80 concurrent agents on a single B200 with P95 end-to-end TTFAT <600ms. Including network overhead. That's an LLM cost-per-minute around $0.0025. (1/4 of a cent.) At a latency lower than any third-party API offers today.
More details about this model, including weights on @huggingface, how to spin it up with one click on Modal, and a starter project repo you can clone, are in the thread ...
Aleix Conchillo Flaqué retweeted
Congrats to @DeepgramAI on the GA launch of Flux TTS, now live in Pipecat 🎉
@JonPTaylor looks at how Flux TTS delivers a more consistent voice experience. Flux reads the whole conversation, not just the next line — adaptive tone, consistent pronunciation, clean interruption handling natively (no SSML markup, no style tags!) all at sub-200ms latency.
➡️ docs.pipecat.ai/api-referenc…
Aleix Conchillo Flaqué retweeted
World Cup Voice Commentary right in your MenuBar!
@pipecat_ai powered voice bot, not only reports on the scores but can answer any questions I have about the games!
Using only local models on my M4 Max. Under the hood:
+ Kokoro TTS for commentary
+ Gemma 4e4 (with and without thinking)
+ Nemotron 3.5 for ASR
Built with Pipecat subagents: when the menubar starts, it spins up a main loop of
[STT]
|
[LLM] <-> [subagent when a game starts]
|
[TTS]
There is one subagent per game. Schedules are pulled regularly, so I can ask stuff like “Who will the US play in the round of 32?” and have it reason over fixtures, for example.
A collector ingests live commentary (per match-id, home/away, and metadata) into a local DB and exposes it via a CLI so pipecat's subagents can query it.
Each game subagent gets an initial prompt about “its” match and tools including an /analyze-game-skill. The skill was built and tested with the early R1 games, especially the chaos of Canada-Qatar game (with injuries, goals, and two yellow cards becoming a red!)
The result: when something happens in the game, the TTS chimes in, excited or dejected, about missed chances, corners, cards, etc. Subagents maintain their own context, so with push-to-talk, which invokes the STT, I can ask about what is happening in a current, past, or future games, fixtures, etc.
The main LLM routes to the correct subagent, or can directly check via CLI, or spins up a new subagent to go hunt down answers (e.g., scores about R1 games, schedule changes, etc.)
Today’s rain delays accidentally stress-tested having multiple concurrent games in Round 3, and the subagent orchestration held up nicely.
Clips + screenshot below 👇
Halftime announcement, second goal, and the third goal.
These were voice memo recordings, aargh
Aleix Conchillo Flaqué retweeted
"Every piece of software is going to have a voice interface. Everyone is going to have a voice as a primary input. I just don't think that idea has gotten the traction it deserves yet."
This week on Skywatch, @getbluejay_ai's car podcast, I sat down with @kwindla, co-founder of @pipecat_ai and @trydaily.
Kwin has been betting on real time infrastructure before most people knew they needed it. Ten out of twelve investors passed on Daily in 2016 saying they were just happy making phone calls. That same stubbornness eventually built Pipecat, one of the most widely used open source framework for voice agents. NVIDIA, AWS, and some of the largest companies in the world have standardized on it.
Before all of that, he was building gestural interfaces that ended up in Minority Report. Steven Spielberg shot demo reels for his company. Then came room sized systems for Fortune 500 companies. Then the bet nobody else wanted to take. A few things from this conversation I have not stopped thinking about:
- Almost every voice agent deployed in production today runs on a foundation model from one of the big labs. Kwin thinks that will not be true in two or three years. The future is open, customizable, low latency models built specifically for real time use cases.
- During early COVID when video usage exploded overnight, Kwin did not sleep more than three hours at a time for three months. They kept adding servers and chasing the next bottleneck. A lot of that growth turned out not to be durable revenue. He kept serving everyone anyway. That is the kind of thing that builds real infrastructure companies.
- Continual improvement is still unsolved. Closing the loop so your agents get better automatically from production data without human intervention at every step is what Kwin is calling the defining engineering challenge of 2026.
If you are building in voice AI, real time infrastructure, or agentic systems, this one is worth your time.
Full episode on YouTube and Spotify in the comments 👇
Aleix Conchillo Flaqué retweeted
Voice AI Meetup, Thursday May 7th. This one's a special crossover event. T-Bot, who hosts the global Voice AI Spaces meetups, is visiting San Francisco and will MC!
- NVIDIA researchers will present some of their really cool recent work on speech models.
- We'll have demos and two fireside chats, featuring new developments in models and evals, with @GradiumAI, @ArtificialAnlys, @ServiceNow, and @pipecat_ai.
- And, of course, 🍕 and great conversation.
- Thanks to the @trychroma team for hosting in their wonderful office/event space.
Registration link below. Come hang out with 150 old and new friends!
🤖 Made with AI
Aleix Conchillo Flaqué retweeted
Sub-agents in (latent) space!
We’ve been working on a side project.
As far as I know, this is the first massively multiplayer, completely LLM-driven game. Come play Gradient Bang with us. See if you can catch me on the leaderboard.
This whole thing started because I wanted to explore a bunch of things I’m currently obsessed with, in an application of non-trivial size, that felt both new and old at the same time.
So … a retro-style space trading game built entirely around interacting with and managing multiple LLMs. Factorio, but instead of clicking, you cajole your ship AI into tasking other AIs to do things for you.
Some of the things we’ve been thinking about as we hack on Gradient Bang:
- Sub-agent orchestration
- Partial context sharing between multiple LLM inference loops
- Managing very long contexts, and episodic memory across user sessions
- World events and large volumes of structured data input as part of human/agent conversations
- Dynamic user interfaces, driven/created on the fly by LLMs
- And, of course, voice as primary input
If you’ve been building coding harnesses, or writing Open Claw agents, or doing pretty much anything that pushes the boundaries of AI-native development these days, you’re probably thinking about these things too!
This is all built with @pipecat_ai, the back end is @supabase, the React front end is deployed to @vercel, and all the code is open source.
Sub-agents in (latent) space!
We’ve been working on a side project.
As far as I know, this is the first massively multiplayer, completely LLM-driven game. Come play Gradient Bang with us. See if you can catch me on the leaderboard.
This whole thing started because I wanted to explore a bunch of things I’m currently obsessed with, in an application of non-trivial size, that felt both new and old at the same time.
So … a retro-style space trading game built entirely around interacting with and managing multiple LLMs. Factorio, but instead of clicking, you cajole your ship AI into tasking other AIs to do things for you.
Some of the things we’ve been thinking about as we hack on Gradient Bang:
- Sub-agent orchestration
- Partial context sharing between multiple LLM inference loops
- Managing very long contexts, and episodic memory across user sessions
- World events and large volumes of structured data input as part of human/agent conversations
- Dynamic user interfaces, driven/created on the fly by LLMs
- And, of course, voice as primary input
If you’ve been building coding harnesses, or writing Open Claw agents, or doing pretty much anything that pushes the boundaries of AI-native development these days, you’re probably thinking about these things too!
This is all built with @pipecat_ai, the back end is @supabase, the React front end is deployed to @vercel, and all the code is open source.
Aleix Conchillo Flaqué retweeted
Join us on Thursday in SF for conversations about voice agents, speech models, and realtime AI infrastructure.
I'm on a panel with:
- @natrugrats from @DeepgramAI
- @farazmsiddiqi from @getbluejay_ai
- Aaron Lee from Parakeet Health
There will be food and lots of opportunities to ask questions and share your knowledge.
One thing I'm looking forward to is comparing notes about GTC last week.
Aleix Conchillo Flaqué retweeted
NVIDIA Nemotron 3 Super launches today! We've been building voice agents with Super's pre-release checkpoints and running all our various tests and benchmarks.
Nemotron 3 Super matches both GPT-5.4 and GPT-4.1 in tool calling and instruction following performance on our realtime conversation, long context, real-world benchmarks. GPT-4.1 is the most widely used LLM today for production voice agents. So an open model that performs as well as GPT-4.1 on hard, voice-specific benchmarks is a big deal.
(Side note: we don't think a benchmark "tells the story" about a model's voice agent performance unless it tests model correctness across at least 20 human/agent conversation turns.)
The Nemotron models are *fully* open: weights, data sets, training code, inference code.
Nemotron 3 Super is 120B params, with a hybrid Mamba-Transformer MoE architecture for efficient inference. You can run it on NVIDIA data center hardware or on a DGX Spark mini-desktop machine.
1M token context.
Blog post with full benchmarks, thinking budget notes, inference setup on @Modal, and where we think this goes next. 👇
Hi @Microsoft ! Can you help me recover my son's account? He can't sign-in to Minecraft anymore which (as you can imagine) is a big deal. Still waiting to hear back from account.live.com/acsr. I’d really appreciate any help. Happy to continue via DM.
This is just too fun!
One of my 2026 predictions is that we're going to see a lot of interesting new experiments with LLM-powered games. There are just so, so many possibilities. The main barrier is inference cost. But that's dropping fast.
My friends Vanessa and Sunah have been tinkering with a voice game called Crush Quest.
Crush Quest has multiple characters, a bunch of really good prompting, and you can play on the web or (clone the repo and) wire up a telephone number. It's, you know, totally open source and that's radical.
As you can maybe tell from my hip use of slang, Crush Quest is set in the early 1990s. It's an homage to a classic electronic board game called Dream Phone. Check out the thread below for a link to the most perfectly 1991 TV commercial for Dream Phone. I can taste the Lucky Charms when I watch this commercial.
h/t to @chelcietay who I had a great conversation with recently about our 2026 predictions and where social and gaming is going.
Aleix Conchillo Flaqué retweeted
Voice-controlled UI.
This is an agent design pattern I'm calling EPIC, "explicit prompting for implicit coordination." Feel free to suggest a better name. :-)
In the video, I'm navigating around a map, conversationally, pulling in information dynamically from tool calls and realtime streamed events.
There are two separate agents (inference loops) here: a voice agent and a UI control agent. They know about each other (at the prompt level) but they work independently.
Aleix Conchillo Flaqué retweeted
Benchmarking LLMs for voice agent use cases. New open source repo, along with a deep dive into how we think about measuring LLM performance.
The headline results:
- The newest SOTA models are all *really* good, but too slow for production voice agents. GPT-4.1 and Gemini 2.5 Flash are still the most widely used models in production. The benchmark shows why.
- Ultravox 0.7 shows that it's possible to close the "intelligence gap" between speech-to-speech models and text-mode LLMs. This is a big deal!
- Open weights models are climbing up the capability curve. Nemotron 3 Nano is almost as capable as GPT-4o. (And achieves this with only 30B parameters.) GPT-4o was the most widely used model for voice agents until quite recently, so a small open weights model scoring this well is a strong indication that production use of open weights models will grow this year.
Voice agents are a moderately "out of distribution" use case for all of our SOTA LLMs today. Literally, in the sense that there's not enough long, multi-turn conversation data in the training sets.
Everyone who builds voice agents knows this intuitively, from doing lots of manual testing. (Vibes-based evals!) This benchmark scores LLMs quantitatively on instruction following, tool calling, and knowledge retrieval in long-context, multi-turn conversations.
Pipecat MCP Server now works with local STT (Whisper) and TTS (Kokoro) models! Happy talking! @pipecat_ai
github.com/pipecat-ai/pipeca…
Aleix Conchillo Flaqué retweeted
Voice-only programming with Claude Code ...
I've been playing with @aconchillo's MCP server that lets you talk to Claude Code from anywhere, today. I always have multiple Claudes running, and I often want to check in on them when I'm not in front of a computer.
Here's a video of Claude doing some front-end web testing, hitting an issue and getting input from me, and then reporting that the test passed.
In the video the Pipecat bot is using Deepgram for transcription and Cartesia for the voice. (Note: I sped up the web testing clickety-click sections of the video.)
The code for the MCP server and the Claude skill are in the repo and Aleix wrote a really good README.md.
You can use any of Pipecat's network transports: generally WebRTC, but you could set this up so you can call Claude on the phone if you wanted to.
There's screen capture support, too, so you can view the Claude code window remotely. That's still a little experimental.
Because this is an MCP server, it's not specific to Claude Code. Try it in other environments! It should work in Clawdbot, Codex, etc ...