Resume feature for dealing with rate limits

[KeithHanson]

Sorry for the two issues in a single day here 😅 Again, many thanks for the tool!

One way to improve the experience and an addition I would absolutely love is to allow this project to be able to "pause" while waiting on your subscriptions token rate limits reset, or be able to halt and remember where it left off.

I am about to switch from my claude sub (work provided) to my gpt pro sub (personal) and re-run it, but that's a lot of tokens to burn again.

One could string together a few bash commands and batch it out, so the blast radius from a rate limit is less, but it would be a great addition to the current functionality.

Ideally, one could set how "stale" a recent result must be to consider being re-scanned.

Apologies if this is already somewhere in the project and I have just missed it!

[manticore-projects]

Thank you for the suggestion @KeithHanson!
Although I am not fully sure to understand it well: AURSCAN is meant for installing one single particular AUR package (under the assumption, that AUR packages are the last resort and should be kept at a bare minimum anyways). Scanning this one package is around 1 US cent right now. It should never be the straw breaking the camel's back.

If I missed something here, then please elaborate your use case and/or feel most welcome to provide a PR.

Cheers and best!

[KeithHanson]

@manticore-projects Ah - assuming I want to spend it in api credits. I am using it with a claude subscription, which has token limits every 5 hours and every week.

I was also running: aurscan --update-check

And I have quite a few out of date package (thankfully, in this case).

If there were some local storage of what --update-check is doing, this would be no big deal, and we could skip those that I never updated before it was done.

I'll drop a log of what happens when I hit a rate limit with the subscription in just a bit, but hopefully this helps provide my context.

[manticore-projects]

My honest first thought was: just wait until your limit is freed again? You can simply skip updating/installing AUR packages when the limits are exhausted?
Nevertheless, I am more than willing to incorporate a more sophisticated solution -- I just don't see it clearly yet.

[GeorgelPreput]

I also find myself with quite a few AUR packages installed, and updating quite regularly. Would this token limit issue maybe also be solved by having fallbacks across backends? Suppose one could fall back on OpenAI or a local LLM, but want to first try using Claude with the existing subscription.

My proposal would be to change the level at which fallback is done. For example, you now have AURSCAN_BACKEND which could be "claude" or "openai", for example. For OpenAI, you support AURSCAN_OPENAI_URL_FALLBACK, but that's only for the OpenAI-like API.

Speaking merely in terms of configuration to paint a picture, would you be open to having something like this:

Variable	Options
AURSCAN_BACKEND	claude · codex · api · openai · /path/to/cmd
AURSCAN_MODEL	<model-name> ← this would go in as the model for whichever backend (api / codex / openai)
AURSCAN_URL	optional https://your-api-url-here to override the endpoint, useful also for Claude API if you go through Azure or Amazon Bedrock
AURSCAN_API_KEY	optional sk_1234abcd... key to be used where using pay-per-token / LiteLLM / etc.
AURSCAN_FALLBACK_BACKEND	same options as AURSCAN_BACKEND, used when the original call fails
AURSCAN_FALLBACK_MODEL	same as AURSCAN_MODEL, used when the original call fails
AURSCAN_FALLBACK_URL	same as AURSCAN_URL, used when the original call fails
AURSCAN_FALLBACK_API_KEY	same as AURSCAN_API_KEY, used when the original call fails

Happy to also contribute a (Claude-generated) PR, if you're open for it. I don't know Go, unfortunately

[manticore-projects]

@GeorgelPreput

The only way I see was to create priority ordered config files which would be tried through until one works. Like

~/.config/aurscan/llm1.conf
~/.config/aurscan/llm2.conf
~/.config/aurscan/llm3.conf
~/.config/aurscan/llm4.conf

(Of course you can change names and/or make it sections in just one file or whatever.)

I don't really see the use-case for myself and to me it looks like complexity at very little value added. But if you would like to provide a reasonable PR which does not disturb standard operations, I would not mind at all.

[GeorgelPreput]

@manticore-projects The PR is at #36, reviews very welcome. It's a Claude Code implementation though, since I'm not familiar enough with Go as a language.

Thanks for being open to this code change

[manticore-projects]

Please try v0.6.2 and feed back, it should be solved now.

[GeorgelPreput]

Installed it and it works quite well (on my machine). Using it with paru, it switches nicely from one back-end to the other.
Thank you!

[manticore-projects]

@GeorgelPreput: I have to thank you since you provided this implementation. Glad it works! best and cheers!