Skip to main content
Point a Mastra Agent at the VLM Run OpenAI-compatible chat endpoint. Mastra runs your local tools; Orion-2 runs document, image, audio, and video work in its code-execution sandbox.

Quick start

providerOptions.vlmrun is merged into the chat-completions body (provider name must match createOpenAICompatible({ name: "vlmrun" })).
Use vlmrun-orion-2:auto for the default tier. Other variants are listed in Code Execution. Get an API key from API Keys.

Examples

Public sample files — same agent and runOptions() for each. Prefer agent.stream (some VLM Run keys only allow stream=True).

Document

Image

Audio

Video

Hybrid: extract + local ERP tool

What runs where

Code Execution

Orion-2 sandbox, libraries, and model variants

Pydantic AI Compatibility

The same hybrid contract from a Python agent

Agent Inputs

Document, image, audio, and video content parts

Artifacts

Retrieve annotated images, clips, and generated files