LM Studio Bionic adds Canvas and coexistence with LM Studio in three releases ending October 12

LM Studio released version 1.1.8 of LM Studio Bionic, its agent app for open models, on October 12, 2026, the third Bionic update in 20 days. The run began with version 1.1.6 on September 23, which introduced Canvas, described in the changelog as an “interactive board that both you and Bionic can edit”. Version 1.1.7 followed on October 1 with Canvas export to PNG, SVG and editable Excalidraw files. Version 1.1.8 lets Bionic run beside the original LM Studio app and use it for local model inference, but only with LM Studio 0.4.26 or newer, a version the same changelog dates October 13, a day after Bionic 1.1.8. Every claim below comes from LM Studio’s own changelog, blog and documentation, saved on October 15.

Takeaway points

• LM Studio shipped Bionic 1.1.6 on September 23, 1.1.7 on October 1 and 1.1.8 on October 12, 2026, adding Canvas, Canvas export to three file formats and coexistence with LM Studio.
• The coexistence feature needs LM Studio 0.4.26 or newer, and LM Studio’s changelog dates that release October 13, one day after the Bionic version that depends on it.
• LM Studio’s Bionic documentation has no page on Canvas, and we have not run Bionic or any of these three versions.

What is LM Studio Bionic, and how is it different from LM Studio?

LM Studio Bionic is a separate app from LM Studio, and the company describes it as “LM Studio’s agent for open models” on its product page. LM Studio itself is the company’s original desktop app, and the product page says Bionic’s local models are “Powered by the LM Studio runtime”. Bionic is an agent: it takes a task, works through files, code and documents, and asks for approval at set points. The company announced it on its blog on July 16, 2026, under the title “Introducing LM Studio Bionic: the AI agent for open models”.

The documentation calls Bionic “the next generation of LM Studio designed for agentic work with open models.” It does not replace the older app. Both the launch post and the documentation say “Bionic is a new, separate app from LM Studio,” and add that “for advanced low-level configuration, you can continue to use LM Studio alongside Bionic.”

Bionic can run a session on three kinds of model, according to its documentation.

The same page says that runtime has “MLX and llama.cpp under the hood.” For readers who want to know what the MLX half of that means on a Mac, we have a piece on what MLX is in LM Studio. The cloud side is a paid service. The models page says cloud usage “will consume the credits available in your account”, and the launch post commits to “Zero Data Retention and never training on your data.” The product page names GLM 5.2, Kimi K3 and DeepSeek V4 Pro as models for the heaviest tasks.

Bionic has been updated often since launch. By our count of the changelog, LM Studio published nine Bionic versions between 1.1.0 on August 27 and 1.1.8 on October 12.

What does Canvas do in LM Studio Bionic 1.1.6 and 1.1.7?

Canvas is a shared drawing board inside Bionic that both the user and the agent can change, according to the 1.1.6 release notes of September 23. That one line is the whole of LM Studio’s published description. The changelog does not say what can be placed on the board, how the agent reads or edits it, or which models can use it.

Version 1.1.7, dated October 1, added the first way to take work off the board. Its notes list “Canvas export to PNG, SVG, and editable Excalidraw files.” The note calls only the Excalidraw files editable; it says nothing about whether a board can be brought back into Bionic after export. Version 1.1.8 added one more Canvas line, a “Draggable Canvas panel”, and nothing else about the feature.

The same three versions changed how Bionic shows and edits files. The table sets the file and session items side by side.

Context compaction is the step an agent takes when a conversation grows too long for the model’s context window: older turns are condensed so the session can continue. LM Studio’s notes give no figure for how much a session is shortened, or when compaction starts.

What changed in LM Studio Bionic 1.1.8 for people who also run LM Studio?

Bionic 1.1.8 can now run at the same time as the LM Studio app, and while LM Studio is open, Bionic hands local model inference to it. The release note reads: “Run both apps together; Bionic uses LM Studio for local model inference while LM Studio is running. Requires LM Studio 0.4.26 and newer.”

The matching note on the LM Studio side, in build 3 of version 0.4.26, describes the same feature from the other end and “Requires Bionic 1.1.8 and newer.” Both apps have to be current for it to work. The LM Studio changelog dates 0.4.26 October 13, 2026, and the release page for it reads “Build 5”. The page gives one date for the version and none for each of its five builds, so it does not show when build 3, the one with the coexistence note, first became available. What the record shows is a Bionic release dated October 12 that names as its requirement an LM Studio release dated October 13.

The two changelogs share several lines word for word. Both list a fix for CUDA not being detected on Windows on Arm, CUDA 13 support on Windows on Arm, a fix to compatibility between Bionic and lms, LM Studio’s command line tool, and the two new API endpoints described below.

They differ on the inference engine. Bionic 1.1.8 lists “llama.cpp 2.53.0 extension packs”, having also listed 2.51.0 and 2.52.0 packs in the same release. LM Studio 0.4.26 build 2 says it “Updated the llama.cpp runtime to 2.50.0 (upstream b11337).” Neither page says which engine version handles a request when Bionic passes local inference to LM Studio.

Memory on a Mac running two model apps is a separate question, covered in our piece on two models at once on a 48 GB Mac. LM Studio does not say whether the coexistence mode avoids loading the same model twice, or how memory is shared between the two apps.

Which API endpoints and decision model controls did Bionic 1.1.8 add?

Bionic 1.1.8 added two API endpoints, /v1/systemone and /v1/decisions, and LM Studio 0.4.26 build 4 lists the same two. The changelog describes the first as a “Jev (TypeSafe AI) compatible endpoint” and the second as an “OpenAI compatible endpoint.” It gives no request format, no example and no link to a reference page for either.

The same Bionic release lists “Decision model load settings, activity status, and cancellation controls” and “Decision model inference and management through the REST API.” The changelog does not define a decision model. Decision models are also the subject of a separate story in this batch, on Ollama adding decision models.

Other items in 1.1.8 are smaller. Skills, the repeatable instructions Bionic added in August, now have their own category in composer suggestions. The release lowers disk use from project history, fixes an incorrect MCP connection status in chats, and translates the native menus and approval prompts into the user’s chosen language.

What has LM Studio not published about Bionic Canvas and coexistence?

LM Studio has published no documentation page, blog post or guide for Canvas. The Bionic documentation we saved on October 15 has sections on projects and sessions, billing, documents, coding, skills, models and voice input, and none on Canvas. The blog’s most recent Bionic posts, as saved the same day, cover the Splash engine, session references, Auto Review, skills and individual models. None covers Canvas, coexistence or the new endpoints.

Several facts a reader would need are missing from the record:

• Whether Canvas works with local models, cloud models or both, and whether a local model needs particular tool support to edit it.
• How the agent sees the board, and whether Canvas content counts towards the session’s context.
• What happens to a Bionic session that is using LM Studio for inference when LM Studio is closed partway through.
• The request and response formats for /v1/systemone and /v1/decisions.
• Which llama.cpp version runs a local request under coexistence.

The date order of the two releases is a point LM Studio could settle by publishing a date for each LM Studio build. Until then the changelog, read as published, has Bionic 1.1.8 requiring a version of LM Studio that is dated after it.

We have not run Bionic 1.1.6, 1.1.7 or 1.1.8, or LM Studio 0.4.26, on our Apple M4 Pro test machine with 48 GB of memory, so this piece has no speed, memory or quality figure of our own. Other agent apps are compared in our piece Pi vs OpenCode vs Hermes on the same local model.