TEN Framework

Open-source framework for conversational AI 🚀
Github:

[TEN Framework @TenFramework]

Physical AI is becoming real — not just models, but devices that can hear, speak, and interact with the world.

During #GTC2026 we're hosting a full-day meetup + workshop with builders exploring Voice AI and Physical AI hardware.

Bonus: 40 dev kits available on site 👀

Bay Area devs — come build with us.

[RTE Dev Community @rtedevcommunity]

🔥 Physical AI Day! Bay Area Devs, see you on March 19!

Happening during GTC week, join RTE Dev Community& @TenFramework for a full day of brainstorming and building with two hardcore events seamlessly hosted at the same venue:

🌅 9:30 AM | Meetup: Conversing with the Physical World
Join industry leaders from @AgoraIO @MiniMax_AI @evermind @RiseLink_X, HumanTouch, and Resonance Ventures @karal127 as we dive deep into the opportunities and future of Multi-Modal & Edge AI.

🛠 1:30 PM | Workshop: Hands-on Voice AI Hardware
Build and deploy a voice AI Agent using the TEN Framework. Here’s the best part 👉 We’re providing 40 Agora R1 dev kits on-site. Successfully run your code, and you get to take the hardware home for FREE!

💡 Note: Morning and afternoon sessions require separate registrations. Spots are highly limited, so act fast! Links in the thread 👇

#PhysicalAI #EdgeAI #VoiceAI #GTC2026

[TEN Framework @TenFramework]

If you're building real-time voice agents, try changing your VAD.

In LiveCap CLI’s optimization runs, TEN VAD reaches the best final CER on 3/4 Japanese engines, and wins on EN qwen3asr (3.61% WER).

Same ASR, same optimization pipeline.
Different VAD → different performance ceiling.

Try TEN VAD: https://t.co/TEmqVhqeuf

[波海瀬🦊VHub @Mega_Gorilla_]

LiveCap CLIについてVAD最適化タスクを実行し、

日本語: 7.06% (CER)
英語: 1.66% (WER)

を達成しました!!

[TEN Framework @TenFramework]

Now live in TEN. Try it free at https://t.co/4AfmGEDjDP

Realtime models are getting better. Voice AI is getting real.

Let’s build!

[OpenAI Developers @OpenAIDevs]

Voice workflows just got stronger with gpt-realtime-1.5 in the Realtime API.

The model offers more reliable instruction following, tool calling, and multilingual accuracy.

Demo with @charlierguo

[TEN Framework @TenFramework]

Happy to join Memory Genesis Competition 2026 as a partner 🚀
Voice agents become truly useful only when they can remember — across conversations, contexts, and time.
Excited to see builders explore long-term memory for real, agentic voice experiences.

[TEN Framework @TenFramework]

Windows + Go + TEN Framework = 🚀

Big news! TEN Framework v0.11.52 is out, officially bringing Go binding support to Windows!

✅ All Go integration tests passed on Windows! ✅ Linux/macOS: C++, Go, Python, Node.js (Full Support) ✅ Windows: C++, Go (Ready to use!)

Next stop: Bringing Python & Node.js to Windows. We’re almost there, stay tuned! 🛠️

See the green lights on GitHub: 🔗 https://t.co/GFPbgPGmfL

[TEN Framework @TenFramework]

Starting the year by building — not talking.

TEN Framework is part of the #2026NewYearChallenge ⚙️
10 days to improve real agent foundations, together with the community.
See you in the PRs 👀
https://t.co/sOKooep8Ew

[memU @memU_ai]

🎉 It’s On! The 2026 New Year Challenge — Registration Is Now Open! ✅

For the next 10 days, developers from around the world will be contributing real code to five projects, tackling practical problems across the agent and AI infrastructure stack.

If you’ve been waiting for a reason to finally build, integrate, or improve something meaningful — this is it.
🌐 Jump in & see all challenges: https://t.co/VRAHWCfyu3

🔧 Choose where you want to build 🔧
Each project runs its own challenge, with independent tasks, rules, and rewards:

🧠 MemU — Improve long-term memory systems for AI agents → PRs, integrations, workflow experiments, and memory optimizations @memU_ai
🤝 OpenAgents — Turn agent collaboration into real, usable systems → Build agent networks, workflows, and practical use cases @OpenAgentsAI
🧩 Sealos — Ship applications that actually run in the cloud → Full-stack apps, infra setups, and agent deployments @Sealos_io
⚙️ TEN Framework — Work on the foundations of agent frameworks → Core fixes, extensions, and framework-level improvements @TenFramework
🚀 Zeabur — Go from idea to live product, fast → Production-ready deployments with one-click shipping @zeaburapp

You can focus on one track or explore multiple. You can work solo. And every meaningful contribution counts where it belongs.
This isn’t about demos or pitch decks — it’s about shipping, learning, and leaving a footprint in real projects.
#2026NewYearChallenge

[TEN Framework @TenFramework]

PR-friendly mode: ON 🎉🎙️
If you’re building (or shipping) voice agents, come help us push TEN forward — bugs, extensions, docs, anything.
https://t.co/SRNZaaLF5l

See you in the PRs 🚀

[memU @memU_ai]

1 day to go🎉
The 2026 New Year Challenge — 5 Projects United Hackathon kicks off tomorrow 🚀
Ten days. Five projects. Win $3,000💰

Task difficulty spans multiple levels — from tests and documentation, to use case design and feature development, and up to advanced challenges. There are also open-ended tracks if you want to explore your own ideas and propose something new.

This year, five open-source projects are united — each with its own focus and challenge track:

🧠 MemU PR Hackathon — Build the memory layer for next-gen AI agents. @memU_ai Fix bugs, add features, build integrations, and improve contributor docs. Every merged PR earns points, recognition, and memU credits.
🤝 OpenAgents PR Hackathon — Build real-world agent networks and practical use cases. @OpenAgentsAI Explore collaborative agent systems and solve real problems with creative agent workflows.
🧩 Sealos Run Wild Hackathon — Ship real, living applications in the cloud. @Sealos_io From full-stack apps on Sealos to memU-powered agents in Devbox — code, deploy, and conquer.
⚙️ TEN Framework PR Hackathon — Contribute to the agent framework layer. @TenFramework Fix bugs, add extensions, and improve documentation to help shape the framework itself.
🚀 Zeabur Ship It Hackathon — From idea to production, fast. @zeaburapp Deploy production-ready apps using Zeabur AI Hub and one-click deployment — ship something real in days, not weeks.

Whether you’re new to these projects or a long-time contributor — backend, infra, workflows, or docs — there’s a clear path to make real impact.
Use today to explore the repos, run things locally, and decide where you want to contribute.

🗓️ We start Jan 8 at 00:00 PT.
😄 Check event details: https://t.co/VRAHWCfyu3
👉 Join us https://t.co/Z43qGBag4z

#2026NewYearChallenge

[TEN Framework @TenFramework]

@fm100 @waylybaye 🙋

[TEN Framework @TenFramework]

Voice AI is moving fast. 🎙️✨

The TEN Framework v0.11.50 is out, featuring a brand new WebSocket Voice Assistant Quickstart. Build + Run in <10 mins.

Everything you need for a production-ready agent:
🟢 ASR: Deepgram / Azure
🟢 LLM: OpenAI (with Tool Calling!)
🟢 TTS: ElevenLabs / Minimax
🟢 Stack: Task, Bun, uv (fastest DX)
🟢 Metrics: Full Grafana/Loki monitoring

It’s like Legos for Voice AI, swap extensions visually via the TMAN Designer without rewriting your core logic. 🧩

Give it a spin:
🔗 https://t.co/4KwG9N8L68
🔗 https://t.co/Jiwq0GjlVR

RTs are highly appreciated! Let's get building. 🚀

[TEN Framework @TenFramework]

TEN Framework just got a lot more Pythonic. 🐍✨

Version 0.11.47 is out, and we’ve fully adopted pyproject.toml. This means better compatibility and a smoother experience for developers building in the Python AI Agent world.

Dive in and give it a try: https://t.co/5fc2pqhyyQ

[TEN Framework @TenFramework]

No more deployment surprises. 🚀

🔒 We’ve just released TEN Framework v0.11.46, and it comes with a small but mighty addition to our CLI: tman install --locked

If you’re familiar with npm ci, you’ll know exactly why this matters.

Using the --locked flag ensures that your installation follows manifest-lock.json strictly. It’s designed to replicate your exact development configuration in your deployment environment.

This means deterministic builds, consistent dependencies, and peace of mind when pushing to production.

Check out the release notes here: https://t.co/snuoJAuvtP

[TEN Framework @TenFramework]

Stop flying blind. Full observability is here for TEN Framework. 📊

In the latest release (v0.11.46), we’ve massively upgraded the runtime metrics for TEN agents. We know how important it is to see exactly what’s happening under the hood, so we didn’t hold back.

We now support: ✅ Prometheus format ✅ OpenTelemetry Protocol (OTLP) ✅ Both Pull & Push models

This makes it incredibly easy to integrate TEN agents into your existing monitoring stacks (like Grafana) to track performance and health in real-time.

This is just the start. The team is actively working on expanding runtime metrics, logs, and traces to give you the granular insights needed to fine-tune your agents perfectly.

Check out the setup guide here: https://t.co/lp3i5IsgW5

[TEN Framework @TenFramework]

🚀 Gemini 3 Flash + TEN Framework = Faster, Smarter Voice Agents! 🚀

TEN Framework is now integrated with Gemini 3 Flash, offering lightning-fast performance and optimized efficiency for your voice agents.

Key Benefits:
⚡️Low Latency: Enjoy fast, real-time responses for smooth interactions.
💰Cost-Effective: Unlock advanced AI capabilities at a fraction of the cost.
🔧Seamless Integration: Easily integrate with TEN’s flexible framework.
📈High-Performance: Perfect for real-time apps like AI assistants and customer support.

Boost your voice agent with Gemini 3 Flash and take your AI to the next level! 👇
https://t.co/H25kqsQTWa

[Google Cloud @googlecloud]

Introducing Gemini 3 Flash—frontier intelligence built for speed at a fraction of the cost.

Learn more about the expanded Gemini 3 model family and how developers and enterprises around the world can get started today ↓ https://t.co/XkKbpSrRao

[TEN Framework @TenFramework]

Thanks for the shoutout! We hope more builders create fun and powerful real-time voice agents with TEN 🙌

[Sumanth @Sumanth_077]

Build production-ready real-time voice AI agents!

TEN is an open-source framework for building multimodal, conversational AI systems that run in real time.

It handles the full real-time loop for voice interactions with audio input, ASR, LLM, TTS, streaming, and tool calls working together at low latency.

You can swap models, customize logic, or assemble pipelines from simple flows to full voice agents.

You can swap components, scale them independently, and also get the flexibility to choose the transport you want, using WebSocket for fast prototyping and RTC for low-latency production.

Key Features:

• Real-time voice pipelines built on a modular actor framework
• Plug-and-play ASR, LLM, TTS, and tool components
• Full multimodal support including text, audio, and vision
• Built-in orchestration for conversational flows and custom logic
• Multi-language support, Python, Node.js, C++, Go

It's 100% Open Source

I’ve shared the link to the guide on building a a Live2D voice assistant with real-time lip sync and the GitHub repo in the comments!

[TEN Framework @TenFramework]

Gradium’s new speech models are seriously good.
We’ve added support in TEN so developers can try them right away.
PR: https://t.co/g11QjZHz5l

[Gradium @GradiumAI]

Gradium is out of stealth to solve voice. We raised $70M and after only 3 months we’re releasing our transcription and synthesis products to power the next generation of voice AI.

[TEN Framework @TenFramework]

Get started: https://t.co/Iz8csON16u

[TEN Framework @TenFramework]

That moment when your voice agent remembers you’re traveling to Australia and that you love potatoes… it suddenly feels way more human 😄

This is TEN × memU real-time memory in action:
it listens, remembers instantly, and uses it in the very next turn.

[TEN Framework @TenFramework]

Love seeing collaborations like this 🙌
Real-time voice + long-term memory is a powerful combo.
Great work from @memU_ai — the integration with TEN Framework is now even smoother ⚡️
If you’re building fast, production-ready voice agents that actually remember, this is worth checking out 👇

[memU @memU_ai]

🚀 We’ve just updated our docs for TEN Framework x memU integration!
🔊 Build real-time voice agents @TenFramework powered by long-term memory with @memU_ai
Docs 👉https://t.co/lXiY4EyFe6

🎙️ End-to-end real-time speech
🧠 Cross-session, contextual memory
⚡ Fast, adaptive, human-like responses

Create voice assistants that truly remember your users, powered by @memU_ai's agentic memory and @TenFramework's low-latency real-time voice architecture.

[TEN Framework @TenFramework]

Community wins like this always make our day 🙌
After proper tuning, TEN VAD improved by +3.8% on JA, becoming the top performer in the test. 😊
Great benchmarking by @Mega_Gorilla_ !
👉 Try it yourself: https://t.co/qrJH8xkE4E

[波海瀬🦊VHub @Mega_Gorilla_]

VADパラメータの最適化結果！
最適化前は、EN,JA問わずjavadが１位だったが、
最適化後は、EN→ webrtc、JA→ TenVADが最適に！

tenvadに至ってはパラメータ修正で3.8%もスコアを上げてる！すげー
(JAvadは仕様の問題で最適化できず...)

[TEN Framework @TenFramework]

https://t.co/qrJH8xkE4E

[TEN Framework @TenFramework]

This week we merged 3 major PRs into TEN VAD — all from the community 😍🔥

✅ Python ONNX inference (Linux/macOS)
✅ Golang support across Linux/macOS/Windows
✅ Java support for Linux/macOS/Windows/Android

TEN VAD is becoming everywhere-ready.
If you’re building a voice agent in 2025, definitely check it out 👇