LM Studio Bionic iconLM Studio Bionic

LM Studio's desktop agent for open models: code and document projects on local models via llama.cpp and MLX, or US-hosted open models with Zero Data Retention

$0 With Local Models
$20 Bionic+ per Month
Jul 2026 Launched

Overview

LM Studio Bionic is a desktop AI agent from the LM Studio team, launched on July 16, 2026 and billed as "the AI agent made for open models." It is a separate app from LM Studio. Work is grouped into projects: a Code project points at a local folder and the agent can search and edit the repository, use Git and run shell commands, with inline diffs; a Work project handles PDFs, decks, spreadsheets, images and notes in a sandbox, with automatic checkpoints for rollback and in-app previews. Models come from three places: local models downloaded inside Bionic and run on the LM Studio runtime (llama.cpp and MLX), models on another of your machines through LM Link, or larger open models such as Kimi K3, GLM 5.3 and DeepSeek V4 Flash on LM Studio's US-hosted cloud, which runs under Zero Data Retention and never trains on your data. It also has a voice keyboard that transcribes on-device with Mistral's Voxtral, Agent Skills in the SKILL.md format, optional web search and MCP servers. Local use is free and needs no account; cloud models need an LM Studio account and a paid plan or credits. The app is closed source, a point Hacker News commenters raised at launch. Also from LM Studio: LM Studio, the local model runtime.

The Verdict

Who Should Use LM Studio Bionic?

Best For

  • LM Studio users who want an agent, not just a chat window
  • Coding and document work that must stay on your own machine
  • Mixing local models with hosted open models in one app

Not Ideal For

  • Machines with under 16 GB of RAM
  • Teams that want frontier closed models such as Claude or GPT
  • Policies that require open-source, auditable code

What's Great

  • A real local-first agent path for open models
  • Clear split between Code and Work projects
  • MLX on Apple Silicon and offline voice with Voxtral
  • Optional cloud with a Zero Data Retention commitment

Watch Out For

  • Coding quality depends almost entirely on the model you load
  • Small local models struggle with multi-file work; 24 GB of RAM runs tight
  • Early UX: thin file-change visibility, less polished than Codex
  • File checkpoints do not undo everything an agent does through the shell
  • Closed source, like the LM Studio app

Pricing

View all features & details

Key Features

  • Code projects: repo search and edits, Git, shell commands, inline diffs
  • Work projects: PDFs, decks, spreadsheets, images, sandboxed
  • Automatic checkpoints and in-app previews
  • Session tabs, side-by-side views, forking a response
  • Agent Skills (SKILL.md), picked with @
  • MCP servers and connected apps
  • On-device voice keyboard with Voxtral, usable in any app

Model Sources

  • Local: GGUF or MLX models on the LM Studio runtime
  • LM Link: models on another device you own, no cloud credits
  • Cloud: hosted open models, Zero Data Retention

Hardware

  • 16 GB RAM is the practical floor for light models
  • 32 GB or more for better coding models plus voice
  • Voxtral voice adds about 3 GB of model weights

Real-World Usage

Launch

  • Announced July 16, 2026 for Mac and Windows
  • Launch coding models named: GLM 5.2 and Kimi K2.7 Code

Reviewer Take

  • Hands-on review: strong local-first idea, early polish
  • Guide advice: pick a model that supports tool use, and check agent results

How It Compares

Feature LM Studio Bionic LM Studio
Job Agent: edits code and documents Run, chat with and serve local models
Models Local, LM Link or hosted open models Local models
Price Free locally; Bionic+ $20, Pro $100 a month Free at home and at work
Source code Closed Closed app, MIT CLI and SDKs

Sources: Bionic · Launch post (Jul 2026) · Pricing · Flavio Copes guide · Bitdoze review · see also Mistral Vibe

Is this your tool? Add a badge