Yttri 0.86.0-beta.1: First Obsidian Plugin, MLX Engine and a Unified Agent Interface

Yttri has released version 0.86.0-beta.1, a major beta update with four areas of change: the agent gets a single section in the interface, a plugin platform has launched with a first integration for Obsidian, an alternative local MLX engine now runs on Macs with Apple Silicon, and voice input and recording have been through a dense series of stability patches.

What changes for agent users

Before 0.86.0-beta.1, Yttri's agent features were scattered across different sections of the app. The new version gives the agent a dedicated "home": a single section with a central entry point from which you can launch tasks, browse session history and manage active processes.

The change solves several problems at once. First, users no longer waste time hunting for the right control. Second, a unified interface makes agent mode more visible and lowers the barrier for those who haven't used it regularly. Consolidating agent functionality is a typical step for a maturing product that moves from a set of features to a considered user experience.

Why plugins arrived, and what Obsidian has to do with it

The release's main systemic news is the launch of an official plugin platform. Until now, Yttri was a closed product with no extension points for outside developers. They now have an official architecture for building integrations.

The first plugin is an integration with Obsidian, a popular tool for keeping a personal knowledge base built on local Markdown files. Key details of the update:

• Release: 0.86.0-beta.1 (beta version)
• Plugin platform: open, and the first plugin is already available
• Obsidian integration: direct access to the local vault without the cloud
• MLX: an alternative local engine for Macs with Apple Silicon
• Voice input and recording: several waves of stability patches

The Obsidian integration lets the Yttri agent work directly with a user's local notes without uploading them to the cloud. For people who keep work archives, research databases or personal knowledge systems in Obsidian, this opens up new scenarios: the agent can build answers using the accumulated context rather than being limited to external sources.

Choosing Obsidian as the first plugin was no accident. Obsidian's audience overlaps heavily with people who work seriously with AI tools: developers, researchers and content creators. Launching the platform with this particular integration points to deliberate positioning.

The MLX engine: native AI for Macs with Apple Silicon

Users of Macs with M1, M2, M3 and M4 chips now get support for MLX, the open-source machine learning framework Apple released in December 2023. MLX is designed specifically for Apple Silicon's unified memory architecture: it uses the CPU, GPU and Neural Engine through a single pool of RAM, which reduces latency when data moves between components.

In practice, this means faster local inference than CPU-only approaches. The data stays on the device, with no upload to the cloud. MLX has been added to Yttri as an alternative to other local engines: users pick whichever works best in their particular configuration.

Voice input gets more stable

The voice features, speech input and recording, have gone through a "dense series of fixes" for stability. In earlier versions, voice mode was among Yttri's least reliable components. In this beta cycle, the developers have clearly made it a priority. The release notes don't spell out the specific fixes, but systematic work on voice signals that the feature is moving toward production quality rather than staying experimental.

What this means

Yttri 0.86.0-beta.1 marks a shift from a set of features to a product platform. Plugins open an ecosystem to third-party developers, MLX covers an important segment of the Mac audience, and the consolidated agent interface prepares the product for a wider circle of users. Which plugins arrive next will show where the team sees the tool heading.