* This update streamlines the repository by removing five obsolete one-off migration scripts (`add_progress.py`, `apply_all_fixes.py`, `fix_step1.py`, `insert_cache.py`, `update_ui_final.py`), having fully absorbed their logic into `app.py`. The core enhancement involves a robust VLM caching system in `app.py` that introduces a `_vlm_merged_cache` dictionary keyed by the hash of the original uploaded screenshot, enabling immediate retrieval of full analysis results on subsequent "Analyze" clicks without re-running the 60–90 second VLM process. This implementation was validated with five new Test-Driven Development tests in `tests/test_vlm_cache.py` (bringing the total to 10 passing tests), confirming that cache hits return instantly while new images correctly trigger fresh analyses, leaving a clean root directory containing only essential application files.
* add example
Inference (#17)
Carlos Salgado
e9a4841
* add modal deploy script and test
* Update Modal deployment to MiniCPM-V-4.6, add HF_TOKEN handling, and add test suite
* fix: regenerate requirements.txt with modal dependency included
The requirements.txt was previously exported before modal>=1.4.3 was
added to pyproject.toml dependencies. Running uv export with both
--extra space and --extra modal includes it now.
* fix: rename ad-hoc Modal test scripts from test_* to check_*
Root-level test_*.py scripts make real network/VLM calls to Modal
endpoints, but are collected by pytest due to the test_ prefix. The
project rule (AGENTS.md) requires all VLM calls to use mock_vlm_factory
in tests. These are debugging/integration scripts, not unit tests, so
renaming to check_* prevents pytest collection while preserving them for
manual use.
* Revert "fix: rename ad-hoc Modal test scripts from test_* to check_*"
This reverts commit 757efd04c77dd3f20aa051f32ef3b96b08b7efff.
* fix: move ad-hoc Modal test scripts into tests/ with env-base…
