Very cool to see Gemma 4 12B.
We’ve tried #Bub with Gemma 4 E2B for simple on-device agentic workflows — so a stronger multimodal model for laptops and edge devices is an exciting step.
Local agents keep getting more practical.
Meet Gemma 4 12B!
A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license.
Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇
😜 Better observability: export data from Tape to @ArizePhoenix without any additional instrumentation— just install an otel plugin. Plus, #Bub and #Deepseek V4 Pro work seamlessly.
Tutorial 6 is live: use Bub as an #ACP agent.
https://t.co/GizJOIVQIa
Expose Bub through #agentclientprotocol, start with @zeddotdev , and keep models, tools, skills, plugins, and tapes on the Bub side.
#AIAgents#OpenSource
Tutorial 4 is the wild one.
Your prompt in Go. Your llm provider in Rust. #Bub still handles the rest.
We kept Bub 's orchestration in Python. All through #WebAssembly + @extism .
Read it: https://t.co/d84xtFR57i
#harness#agent
recommend reading. LangChain has been showing how to make agents self-evaluate and self-evolve.
@bubdotbuild did something similar with https://t.co/DnuakWxEJR . It’s cool to see that in agent engineering, we’re starting to land on some shared ideas and best practices.
Tutorial: local Agents with #Bub + llama.cpp.
Whether a small local model like Gemma 4 by @googlegemma drive simple agentic workflows?
Not perfect, but surprisingly usable — and Bub’s abstraction makes it easy to experiment.
https://t.co/2P0MoFWDTo
#Agent#LocalAI
Tutorial 3 is live: #Bub with a SQLAlchemy-backed tape store.
Tape keeps the Agent model simple: append-only records for context, memory, and ops logs — whether you start with JSONL/Markdown or move to a queryable DB like #sqlite
https://t.co/lGrGukj4lB
#python#agent#llm
Part 2 of my tutorial series is LIVE!
Learn to connect ANY MCP server to Bub in minutes using bub-mcp. No more struggling with configs – just copy, paste, and run.
Get the guide: https://t.co/M2rZpXYAeC
#MCP#Bub#AI#DevTools#Python
With #Bub, you can use familiar observability tools like logfire or jaeger. Of course, you can also just ask Bub what’s going on.
otel exporter powered by @pydantic logfire
Read More 👇
@sama I need a compact model that works well enough (at a lower price), and I truly hope that everyone in the world can enjoy cutting-edge technology as much as possible.
bub just passed 10k downloads on PyPI 🎉
Many users deploy their own forks, but this still feels like a small milestone.
We’re happy to see people extend, adapt, and ship bub in their own ways.
https://t.co/MEHeWdRKvl
Data screenshot via ClickPy by @ClickHouseDB