Hey there,
Thanks for creating this. I tried it out because it looked cool. A few items as feedback, maybe you would consider implementing some of them (Python is not my world, so for now I won't do any PR's).
1: VS Code Insiders
Maybe consider adding it as an auto-discovered option. I use it mainly (I know, a bit weird) and I had to use the web interface and change the path there. After refreshing it loads the original default path. I'd like it to keep the most recent typed path and even better, better auto-detection or the option to just select my IDE.
2: Legacy JSON files
[13:22:46] Storage path: /Users/myname/Library/Application Support/Code - Insiders/User/workspaceStorage
[13:22:46] Starting scan…
[13:22:46] Discovering session files…
[13:22:46] Found 997 JSONL + 354 legacy JSON files
[13:22:46] Registering workspaces…
[13:22:46] Calculating incremental diff…
[13:22:46] 1 changed, 0 deleted
[13:22:46] Parsing 1 file(s)…
I had quite a lot of legacy files here, and it would be great if they could be parsed, you might be attempting now, but failed for me it looks like. It seems to have gotten stuck here. Six minutes in, still stuck on the same file, and it's unknown the which file it is (maybe explicit output the filename for easier debug). I might have have incomplete data?
3: Haiku at 0.33x
Haiku is at 0.33x, maybe it should also be included with other models at the same pricing.
4: Pricing overall
I think it would be interesting to attempt to estimate costs. Two ways, the official cost per request by Github which should be easy as you know the number of requests and the model used, but more interestingly, estimating the actual API cost per model by using the official pricing for the models and multiplying that by the tokens used. This way I can get an overview over how much I would have paid elsewhere and how good a deal I got from using Github - and we call know it's probably a pretty good deal, but this tool could fairly precisely confirm that I think.
5 Are output tokens correct?
Not sure if 86M input vs 3.9M output is normal?
Hey there,
Thanks for creating this. I tried it out because it looked cool. A few items as feedback, maybe you would consider implementing some of them (Python is not my world, so for now I won't do any PR's).
1: VS Code Insiders
Maybe consider adding it as an auto-discovered option. I use it mainly (I know, a bit weird) and I had to use the web interface and change the path there. After refreshing it loads the original default path. I'd like it to keep the most recent typed path and even better, better auto-detection or the option to just select my IDE.
2: Legacy JSON files
I had quite a lot of legacy files here, and it would be great if they could be parsed, you might be attempting now, but failed for me it looks like. It seems to have gotten stuck here. Six minutes in, still stuck on the same file, and it's unknown the which file it is (maybe explicit output the filename for easier debug). I might have have incomplete data?
3: Haiku at 0.33x
Haiku is at 0.33x, maybe it should also be included with other models at the same pricing.
4: Pricing overall
I think it would be interesting to attempt to estimate costs. Two ways, the official cost per request by Github which should be easy as you know the number of requests and the model used, but more interestingly, estimating the actual API cost per model by using the official pricing for the models and multiplying that by the tokens used. This way I can get an overview over how much I would have paid elsewhere and how good a deal I got from using Github - and we call know it's probably a pretty good deal, but this tool could fairly precisely confirm that I think.
5 Are output tokens correct?
Not sure if 86M input vs 3.9M output is normal?