The START project includes a lightweight browser-based GUI for generating custom Active Inference curricula without needing to use the command line.
uv run python -m learning.curriculum_creation.generate_curriculum_gui
This launches a local HTTP server (default: http://127.0.0.1:8765) and opens your default browser.
--allow-remote and an explicit authentication tokenThe GUI server’s own code introduces no third-party dependencies beyond the project’s existing pipeline dependencies (which the shared orchestrator already requires); its HTTP/UI layer uses only the Python standard library. It:
CurriculumOrchestrator from generate_custom_curriculum.py/status endpoint every second for live updatesThe GUI runs all five stages of the pipeline:
| Argument | Description | Default |
|---|---|---|
--host |
Server host address | 127.0.0.1 |
--port |
Server port | 8765 |
--no-browser |
Don’t auto-open browser | False |
--allow-remote |
Opt into non-loopback binding | False |
--auth-token |
Require this header for remote requests | unset |
Example:
uv run python -m learning.curriculum_creation.generate_curriculum_gui --port 9000 --no-browser
The GUI server exposes these JSON endpoints:
| Endpoint | Method | Description |
|---|---|---|
/ |
GET | Serves the HTML GUI |
/options |
GET | Returns available domains, entities, languages as JSON |
/status |
GET | Returns current pipeline progress, completion, error, and run-lock state as JSON |
/results |
GET | Returns an escaped HTML summary of completed run |
/start |
POST | Starts a pipeline run with form data |
/cancel |
POST | Requests cancellation of the active run |
PERPLEXITY_API_KEY, OPENROUTER_API_KEY)/start rejects malformed JSON, non-string fields, oversized fields, and unsafe path-like values before starting a worker/cancel for cooperative cancellation; shutdown also signals the worker cancellation event