cellmap_flow.ai_annotate

AI-assisted annotation: a hosted model proposes a mask for one plane, the user reviews it.

In the Finetune tab the user points at a structure and asks for it to be annotated; the dashboard sends that plane of raw EM to a model with a prompt asking it to paint the structure, turns the reply into a mask, and stages it for the user to accept into the annotation volume or reject.

  • config: the server-side config that turns the feature on and lists the providers and models it may use. Off without it.

  • errors: AIAnnotateError, a category and a message safe to show.

  • secrets: redacting secrets from logs, and reading API keys safely.

  • usage: the per-user daily call limit.

  • images: checking a model’s image before decoding it; PNGs for the browser.

  • organelles, prompts: what the model is told about each structure.

  • mask_extraction: EM plane to RGB, and the model’s reply to a mask.

  • backends: the providers (Vertex Gemini, a local fake) behind one interface.

  • geometry, pipeline, staging, audit: which plane to read and where its mask goes, staging the result for review, and the audit log.

Import the submodule you need; this package imports none of them, so the SDKs a provider needs load only when a call is made.

Submodules