AI & MCP

Oraphim separates two ideas that are often mixed together: focused local AI tools and MCP-connected assistants/agents.

Local AI handles specific media tasks such as Magic Mask, Content Aware Fill, and speech-to-text through native ONNX models. MCP lets a compatible external assistant inspect or author Oraphim projects through a structured, permission-controlled tool surface.

MCP authoring

The certified production MCP registry exposes more than two hundred editor tools covering project inspection, timelines, compositions, nodes, color operations, export/resources, workflow calls, diagnostics, and selected UI-control operations. Tools operate on Oraphim's editor model rather than requiring the agent to guess screen coordinates for normal authoring.

Use ReadOnly for inspection. Grant edit/admin scope only when the assistant needs to mutate the project. Save a milestone before large AI-authored changes and review the resulting project exactly as you would review another editor's work.

Local AI

• EfficientTAM: Magic Mask, smart selection, tracking-oriented segmentation, and object-removal masks.
• LaMa: Content Aware Fill from a source frame plus active mask.
• Whisper tiny.en ONNX: speech-to-text/caption starting point.

These model packages are downloaded on demand and run through Oraphim's native ONNX Runtime path. They do not require a subscription to a general-purpose chat model.

Start here

1. Read Local AI models if you only need masking, fill, or speech tools.
2. Read Connect an AI/MCP client to connect Codex, Claude Desktop-style clients, or another compatible MCP client.
3. Read MCP permissions & security before granting mutation access.
4. Use MCP tool reference when building automation or agent integrations.

AI assistance is still editing: inspect the result, use undo/versioning, save, reopen, and render before approving final work.