Migrating from dump_models.py¶

vexy-co-model-catalog replaces dump_models.py, a single script that fetched model lists from the same providers. A copy of that script is kept in the repository as external/dump_models.py for reference.

What stays the same¶

Provider names. The provider table is the same list of names, types, key variables and base-URL variables.
Environment variables. Keys are read from the same variables (OPENAI_API_KEY, GROQ_API_KEY, HUGGINGFACEHUB_API_TOKEN, FRIENDLI_TOKEN, …), and base URLs from the same <NAME>_API_OPENAI variables.
File names. Output files are still called models_<provider>.json and models_<provider>.txt.
Re-running. Both tools skip a provider whose files already exist, unless you pass --force.

What changes¶

	`dump_models.py`	`vmc`
Where files go	The current directory, flat	`config/json/` and `config/txt/` (or `models/json/` and `models/text/` with `--legacy_output`)
TOML output	`models_<provider>.toml` next to the others	Not written
Failure record	`failed_models.json` in the current directory	`models/extra/failed_providers.json`
Skip check	JSON, TXT and TOML files must all exist	JSON and TXT files must both exist
Retrying failed providers	`--failed_rescan`	`vmc clean --failed`, then fetch again
Concurrency limit	—	`--max_concurrency`, default 8

Migrate step by step¶

Prerequisites¶

vmc is installed. See Installation.
Your API key variables are exported in the shell you use.

Steps¶

Change to the directory that holds your old models_*.json files.
```
cd ~/models
```
Optional: run the migration assistant. It needs a source checkout of the repository, because the assistant script is not part of the installed package.
```
python /path/to/vexy-co-model-catalog/scripts/migrate_from_dump_models.py
```
The assistant looks for models_*.json, models_*.txt and models_*.toml in ., models, ../models and ~/models. It prints how many providers and models it found, which of nine common key variables are set, and a suggested fetch command. It changes no files.
Fetch fresh catalogs.
```
vmc fetch
```
Compare the new files with the old ones. Because the new files are sorted, sort the old ones the same way before you diff, or compare the TXT lists only:
```
diff <(sort models_openai.txt) <(sort config/txt/models_openai.txt)
```
Point your other tools at the new files. If they need the files flat in one directory, create links there:
```
vmc link --target_dir=.
```
This links ./models_<provider>.json and ./models_<provider>.txt to the files in config/. Existing files with those names are replaced by the links, so keep a copy of the old files first if you still need them.
When the new files are in use, archive dump_models.py and the old failed_models.json.

The `migrate` command¶

vmc migrate runs the same assistant script. It looks for the script relative to the installed package, which only works when you run from a source checkout; an installed copy prints Migration script not found. The --search_paths argument is passed to the script as the SEARCH_PATHS variable, but the script does not read it in this version, so it always searches the four default locations listed above.