I see more and more smart people in this community talk about using Claude as an app. I’m using Claude as my primary tool, but I pay for only the tokens used. With two people we used $15-20 over the course of a few months, even with the little bit of coding I’ve been doing again lately.
I use it through BoltAI a chat interface (available via SetApp), Zed a modern development platform and occasionally via DevonThink and ObsidianCopilot.
What is that the Claude app, provides that justifies $20/mth? (As far as I can it won’t support the DevonThink or Obsidian uses)
That I didn’t know. I’ve used OpenRouter for credits, so I can choose which AI I’m using, or compare results. My usage is generally far below a monthly fee and certainly much lower than subscribing to them all.
I’ve recently received a one year academic subscription to Gemini for free… its results are variable! I like Claude’s style…
One question that came to mind was: what about security? When I’m using a third party service to route my questions/ideas/conversations to an AI system it moves through their servers and infrastructure, right?
Given the potential personal or sensitive data that some people talk about with AI systems nowadays I wouldn’t want another player in between that I would have to „monitor“ its security standards and trust them as well besides the actual AI provider. Or am I getting something wrong here?
What subscription do you have that comes with API credits? Mine does not—at least not for general use. I get a capped allotment for Claude Code, but not for anything else. I have to buy API credits separately if I want to use them for, say, Devonthink.
Security and Privacy - the whole GenAI space is a minefield here.
BoltAI - they connect directly to the Al provider from your computer, no middleman, so no increased risk. Open Router - no idea, it could be very secure but I’ve never used it.
Model Provider - OpenAI, Gemini and likely Anthropic are all the real risk. OpenAI stores your chat in plain text for months; Gemini stores your chat history for 18 months and presumably Anthropic is similar to OpenAI. I don’t know about if Gemini and Anthropic encrypt the data. I know OpenAI does not, because it came out in a court case where they were subpoenaed to hand over chat history.
We have a corporate account (Claude Team? I think), so I don’t know if its included in a flat sub or if we pay more, that wasn’t really my point. I just meant I can use API where needed if I want to with that account.
It seems like you now have to have a Pro or other paid plan to be able to create an API code. This makes the instructions on the BoltAI site for using Claude outdated. Has anyone upgraded to a paid account, generated an API code, then dropped back to the free and pay per use by API? That’s what I thought could be possible with the approach described in this thread.
I’m not on a monthly subscription, I just have prepaid credits. I would also note that apparently those credits expire after 12 months, for both OpenAI and Anthropic, so don’t top up too much lest it mostly expire.
Thank you for that information. So at least BoltAI doesn’t seem to be kind of a man in the middle and as for the providers it comes down to „pick your poison“ it seems .
You don’t need a paid plan to create an API code. You do need to create a developer account, but you don’t need to be a capital-D developer to do so, of course. Once you’ve created your account, you can get an API code and buy credits. The minimum purchase is in the $5-$10 dollar range, which you can top up on an as-needed basis.
Start on this page. There should also be a chart laying out how much tokens cost based on which Claude model you use, etc.
Followup on this. A recent ATP episode had an ad for Claude Pro and 50% discount for 3 months:
Based on the comments here and the possibility of using Claude Code I decided to sign up for the 3 month trial and see how it goes.
In addition, Teresa Torres (Agile Product focused person) posted a good article about using Claude Code for non-coding tasks:
Also on the subject of cost. If you’re coding with an AI tool using API tokens, you might use a lot more tokens. This article, is what prompted me to work out the math: