LM Studio Bionic: the AI agent for open models

TL;DR Highlight

LM Studio launched Bionic, an agent app for running local and cloud open models.

Who Should Read

Developers who want to run coding and document agents on open models but worry about cost and data leaks

Core Mechanics

Bionic is an agent app built for open models
- It handles coding, research, and document and file work in one app.
- You can pick local models or cloud open-source models.
Coding happens in a Code project
- Point it at a local folder and it investigates, edits, and debugs code.
- Inline diffs show changes, and agentic code search finds relevant files.
Document work runs in a sandbox
- Work projects process documents, PDFs, decks, and sheets in a sandboxed environment.
- Automatic checkpoints let you review or roll back changes.
You choose how models run
- Options are local, LM Link, or LM Studio Secure Cloud.
- Picking the model and compute per task controls cost.
Voice input runs locally too
- The launch voice model is Mistral AI's Voxtral.

Community Reactions

First impressions from an agent user
- A Codex user ran Qwen3.6 35B and said results were what they hoped for.
- They noted it shows Working instead of Loading model and has no way to preload or unload models.
Complaints about scope limits
- One commenter said it is locked to a single directory with no system-wide access, local web search, or ssh.
- They normally use opencode with LM Studio.
Pushback on closed source
- Commenters pointed out both LM Studio and Bionic are closed source, which clashes with the open-model message.
- One argued the agent layer being open matters more since models can be swapped weekly.
Skepticism about the retention promise
- One commenter said the promise could be reversed later through a privacy policy update.
- Another asked whether it applies to the third-party frontier cloud models.

How to Apply

Try a Code project with your existing LM Studio models
- Download Bionic and connect a code folder as a Code project.
- Pick a model you already have and ask it to fix a small bug or explain code.
Verify document work with checkpoints in a Work project
- Hand it meeting notes or spreadsheet cleanup in a Work project.
- Open the checkpoint diff afterward and roll back if needed.
Split tasks between local and cloud and compare cost
- Run simple tasks on local models and heavy ones on open models in Secure Cloud.
- Run the same task in both and record quality and billed cost.
Check license and data requirements before adopting
- Confirm whether your team policy requires open-source tooling.
- Compare the closed source status and cloud data policy to your internal standards.

Try It

copy & run

# Paste into Bionic's Code project (or any agent) — a simplified version of the article's method

Analyze this project as if explaining it to a new hire.
Target feature: {{feature name}}

1. Search for relevant files and list them.
2. Explain the flow step by step from request to response.
3. Do not modify any code yet.

Next step: run the same prompt on a local model and a cloud open model, then compare explanation accuracy and cost.

Terminology

Agent harnessThe framework that gives an LLM tools like file reading, code editing, and search and runs the task loop. If the model is the engine, the harness is the car body.

Zero Data RetentionA policy of not keeping input data on servers after processing a request. Like a restaurant that takes your order but keeps no receipt.

LM LinkA connection option mentioned in the article for running models. The article gives no further detail.

CheckpointA restore point saved each time the agent changes a file. Like a save point in a game, it lets you return to an earlier state.