Bodega One beta.45 is out.
- Desktop control on Windows: asks first, highlights every action, Esc to stop
- Smarter agent: fewer wasted steps, cheaper long runs, steer it mid-reply
- New models: Sonnet/Haiku 5.5, GPT-6.1 Sol, Gemini 3.6 Flash
- Calmer UI, smoother streaming, one-click background updates
- Instruction sets, per-chat memory, safer by default
7/ CLI v0.2.10: a live /diff pane beside the transcript, the agent's questions beside the prompt, --harness, mcp login, plugin export, shell completion. Free, runs on your machine, no account: bodegaone.ai
6/ GPT-5 and GPT-6 run on OpenAI's Responses API now: tools and reasoning together, reasoning summary in the thinking pane. Local models that write tool calls as text get their calls through on strict templates.
5/ OAuth sign-in for hosted MCP servers, plugin updates that show what changes before installing, and your skills export as a plugin. NVIDIA PAIR, Qwen3.8 hosted, Unsloth Desktop, GGUFs from your Hugging Face cache.
4/ Bodega is now an MCP server. Point Claude Code, Cursor, Codex or any MCP client at localhost:1337/mcp for your memories, knowledge base, docs and code map. Five tools, never shell or file access, nothing leaves your machine.
3/ The agent can ask you a question without stopping. It posts a card and keeps working on everything that does not depend on the answer. Answer when you like; if the task finished first, your answer goes out as your next message.
2/ The harness stays attached between turns, so it keeps its own context too. Pick Bodega to switch back; the thread stays one transcript. A local model can spawn a sub-agent on a harness and merge the result back like any delegated task.
Mayimbe retweeted
Bodega One Code truly is the most BYO(bring your own) app of its kind. True agnostic app.
Should we make it open source?
Bodega One v1.0.0-beta.39 is out.
This one's about honesty. When the agent gets cut off mid-task, it now says so, in the app and the CLI — instead of presenting half an answer as a finished one. Long local sessions stop losing your original request when memory gets tight. And routing now does exactly what the settings page says it does.
Also: Model Roles is four pickers instead of eight, attachments stopped tripping false "too long" errors, and we fixed a filter that was silently eating angle brackets from your messages, pasted code keeps its generics now.
Free . bodegaone.ai
Mayimbe retweeted
Two Bodega One products, one clip. That is Bodega One HQ, our ops platform for people and AI staff, running inside a real iPhone frame. The frame is Bodega One Code's new phone preview, shipped in beta.38. We build one inside the other.
Mayimbe retweeted
◢◤
Call of Duty: Modern Warfare 4
ベータ開催記念CP
◥◣
⠀
抽選でベータ早期アクセスコードを配布。
新兵・復帰兵問わず、誰よりも早く前線へ出撃せよ。
🎮応募方法
①フォロー＆リポスト
②DMで抽選結果が届く
📅締切
8/23(日) 23:59
8/22からベータ早期アクセス開始
プレイした感想も募集中！
Available later today on bodega one, Big W to the @Alibaba_Qwen team!
We promised open weights for Qwen3.8. Now, time to meet them! 🎉
⚡ Qwen3.8-27B:
- A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world coding & office workflows.
- 262K native context, easily extendable to 1M tokens via YaRN.
- Built for builders. Highly efficient, high-quality, and licensed under Apache 2.0.
🚀 The open weights for Qwen3.8-2.4T-A95B (Max-level) have also been released recently.
Whether you're shipping lightweight applications with Qwen3.8-27B locally or building agents with Qwen3.8-2.4T-A95B, they're yours now!
Download, deploy, and build something we haven't imagined yet. 👀👇
- Hugging Face：
huggingface.co/collections/Q…
- ModelScope：
modelscope.cn/collections/Qw…
