auralynq-rag / docs /getting-started /troubleshooting.md
MHamdan's picture
Deploy Auralynq RAG (Llama-3.3-70B via HF Inference Providers)
8c1b9fe verified
|
Raw
History Blame Contribute Delete
3.25 kB
# Troubleshooting
## No-Podman / local dev
- **`make setup` fails to resolve deps** β€” make sure you're on Python 3.11 or
3.12 (`python3 --version`); older interpreters aren't supported
(`requires-python = ">=3.11"`).
- **`auralynq: command not found` after `make setup`** β€” activate the venv:
`source .venv/bin/activate`.
- **Answers look extractive / low quality** β€” you're on the offline fallback
(hash embeddings + in-memory store + extractive LLM). This is expected with
no keys/extras installed and verifies pipeline *integrity*, not quality β€”
see the Limitations section in `README.md`. Install `auralynq[embeddings]`
and/or set an LLM provider key for real quality.
- **Web UI can't reach the API** β€” confirm `NEXT_PUBLIC_API_BASE` matches
where `uvicorn` is actually listening (`http://localhost:8000/api` for the
no-Podman dev flow), and that both processes are running.
- **Port already in use** β€” pass `--port` to `uvicorn`/`npm run dev` or stop
whatever else is bound to 8000/3000.
## Podman stack
- **`make runtime-check` can't find a Compose command** β€” install
`podman compose` (Podman v4+) or `podman-compose` separately; Auralynq
resolves whichever is present via `scripts/check_container_runtime.sh`.
- **Cert warning in the browser** β€” expected with the baked-in self-signed
cert; click *Advanced β†’ Proceed*, or follow [server.md](server.md) to bind a
real domain with Let's Encrypt.
- **502 right after `make stack-up`** β€” the web container is still booting;
retry in a few seconds.
- **No answer in chat after a code change** β€” rebuild the image
(`podman build --no-cache`, not `podman-compose build`, which can reuse
stale layers) and fully cycle the stack (`podman-compose down` then
`make stack-up`); hard-refresh the browser.
- **Rootless networking / container DNS issues** β€” `make stack-up`
auto-patches the CNI conflist (CNI 0.4.0 + dnsname) on hosts where rootless
Podman lacks container DNS; no sudo needed. On a netavark host this is a
no-op.
- **Can't reach the stack from another machine** β€” open inbound TCP on
whatever `AURALYNQ_HTTPS_PORT` is set to (default `8443`) in the firewall;
every other port binds to loopback intentionally.
## Both modes
- **Secrets never show up** β€” they live only in `.env` (git-ignored); if a
provider isn't detected, check the exact env var name against
`.env.example` (nested settings use `__`, e.g. `AURALYNQ_LLM__PROVIDER`).
- **Corpus looks stale after deleting files by hand** β€” don't delete `data/`
subfolders directly; use the API's guarded clear flow
(`POST /corpus/clear/preview` then `/corpus/clear/confirm`) so the vector
store, graph, and page cache are cleared together. See
[no-podman.md](no-podman.md#clearing-your-data-safely).
- **Visual grounding shows "unavailable" for a document** β€” it was indexed
before visual grounding metadata existed, or page rendering failed at
ingest. Re-ingest the document, or use
`POST /documents/{id}/render-pages` to re-render pages without a full
reindex.
## Still stuck?
Open an issue with: your OS, Python/Node versions, which mode (no-Podman /
Podman / server), the exact command that failed, and the full error output.