Stories Actors

Containerized Tapis Actors for Stories UI workflows. This repo holds the Python entrypoints and build workflows for:

• Multi-narrative analysis (CPU)
• Semantic bridge (SVO) mapping (CPU)
• Whisper transcription (GPU)

Layout

• entrypoints/ — Python actor stubs; read JSON from stdin, emit JSON to stdout.
• images/nlp/ — Docker context for CPU NLP actors (multi-narrative, semantic bridge).
• images/whisper/ — Docker context for GPU Whisper actor.
• .github/workflows/ — CI to build/push images (GHCR by default).

Building locally

# NLP (default entrypoint: multi_narrative; override ENTRYPOINT_SCRIPT env to switch)
docker build -t stories-nlp-actor:local images/nlp

# Whisper (GPU-capable base recommended)
docker build -t stories-whisper-actor:local images/whisper

Registering actors

Use the scripts copied from the former UI repo (or recreate) to register images on tacc.tapis.io:

export TAPIS_TOKEN=...   # actors:create/admin
export MULTI_NARRATIVE_IMAGE=ghcr.io/yourorg/stories-nlp-actor:latest
export SEMANTIC_BRIDGE_IMAGE=ghcr.io/yourorg/stories-nlp-actor:latest
export WHISPER_IMAGE=ghcr.io/yourorg/stories-whisper-actor:latest
bash register-actors.sh             # all
bash register-actors.sh whisper     # single

CI workflows

Two workflows build/push images to GHCR:

• build-nlp.yml (cpu) -> ghcr.io/<owner>/stories-nlp-actor:latest
• build-whisper.yml (gpu) -> ghcr.io/<owner>/stories-whisper-actor:latest

Set repository secrets if you need an external registry; GHCR works with GITHUB_TOKEN.

Next steps

• Wire the entrypoints to real pipelines (multi-narrative-analysis, semantic-bridge-demo, whisper-container).
• Add model weights/runtime dependencies to the Dockerfiles.
• Update registration env vars in Stories UI to point at the published actor IDs.