Free LLM routes come and go. I run a couple of LiteLLM proxies and got tired of hand-editing configs every time Groq, OpenRouter or Kilo changed their free tier. So I built a scanner that checks which free routes actually answer, publishes the result on f24-sales.com, and this repo pulls that list into your own LiteLLM. Maybe it saves you the same chore.
Currently covered: Cline, Groq, Infron, Kilo, Nous Portal, NVIDIA, OpenCode Zen and OpenRouter. Any OpenAI-compatible /v1/models endpoint can be added.
LiteLLM YAML · .env template · Import guide
git clone https://github.com/f24sales/litellm-free.git ~/litellm-free
cd ~/litellm-free
python3 -m pip install --target ~/.local/share/litellm-free/python -r requirements-import.txt
export PYTHONPATH=~/.local/share/litellm-free/python${PYTHONPATH:+:$PYTHONPATH}
cp env.example .env && cp config.conf_example config.conf && chmod 600 .env config.confPut your gateway keys and your LiteLLM URL + bearer token into .env. The bearer needs model-management rights; a plain chat key won't do.
Then see what would happen:
python3 import_litellm.py api --env-file .env --dry-runHappy with the diff? Drop --dry-run.
Just want the YAML? No keys needed:
python3 import_litellm.py pull --output config.yaml --forceOnly the routes you actually have keys for:
python3 import_litellm.py file --format yaml --output config.yaml --forceWrite into a running LiteLLM:
| Goal | Command |
|---|---|
| Push via HTTPS API into LiteLLM's database | python3 import_litellm.py api --env-file .env |
| Take over routes an earlier free-sync created | … api --env-file .env --adopt-managed-by free-sync |
| Only update metadata on existing routes | … api --env-file .env --patch-managed-by free-sync |
| Write straight into the Postgres inside the LiteLLM container | … sql --env-file .env --container litellm-database |
config.yaml is a symlink to the newest config_DATETIME.yaml. Set LITELLM_FREE_ARCHIVE=1 in config.conf if you want old versions kept under archiv/. Identical content doesn't create a new file.
- Won't touch models owned by another manager unless you explicitly select that legacy manager with
--adopt-managed-by. - Won't delete anything unless you pass
--pruneor setIMPORT_PRUNE=1. - Won't invent a context window. If the gateway doesn't report a limit, the field stays empty.
- Won't add input and output limits together — they're kept separate, the way the clients expect them.
- Won't change your configured defaults or fallbacks. If one of them disappears upstream, the status report tells you.
Container (Podman shown, Docker works the same):
podman build -f Containerfile -t localhost/litellm-free-import:latest .
podman run --rm --read-only --cap-drop=ALL --security-opt=no-new-privileges \
--env-file config.conf --env-file .env localhost/litellm-free-import:latest api --dry-runsystemd user timer, twice a day at 00:10 and 12:10 UTC:
install -d -m 700 ~/.config/litellm-free
install -m 600 .env config.conf ~/.config/litellm-free/
install -d ~/.config/systemd/user
cp deploy/litellm-free-import.service deploy/litellm-free-import.timer ~/.config/systemd/user/
systemctl --user daemon-reload
systemctl --user enable --now litellm-free-import.timerThe unit expects the checkout at ~/litellm-free and the packages in the directory from the quickstart. No venv needed. Details and the container drop-in are in IMPORT.md.
If IMPORT_HOOK_URL and IMPORT_HOOK_BEARER are set, the importer posts the result to that endpoint after LiteLLM has confirmed the new state — import_succeeded or import_failed, with the model diff in the body and no credentials in it.
The published YAML is the import source of truth on every run. With IMPORT_PRUNE=true and IMPORT_ADOPT_MANAGED_BY=free-sync, obsolete legacy routes in the litellm-free group are reconciled too, even when the incoming YAML is unchanged. The pre-success access-sync command receives the exact imported model names in IMPORT_MODEL_NAMES_JSON, so it cannot reintroduce software-only routes from the raw scan before the final catalog-refresh hook.
The OpenClaw plugin on the receiving side refreshes the model lists in OpenCode, Hermes and OpenClaw and sends a Telegram note about what changed. ops/refresh-models.sh does the same reconciliation by hand. How that reload works internally is documented in ops/.
python3 -m pip install --target ~/.local/share/litellm-free/python pytest
python3 -m pytest -q testsThe SQL path has only been exercised against my own LiteLLM/Postgres setup. The API path is what I use daily. Free tiers are promotional by nature — expect routes to vanish between scans; that's the whole reason this thing exists.
Issues and PRs welcome.