#https://wisprflow.ai/use-cases/github
1 messages · Page 1 of 1 (latest)
@dreamy star yeah I love it. But feel like someone must have an OSS solution. I might just subscribe for a few months while I find an alternative
It's just really one of those products that's so good I tell everybody I know about it
@left field that looks really nice. Oddly similar to Wispr including the logo lol. Are you affiliated with that rec?
Yeah I ran out of free credit and used the Apple speech to text and it was so much worse than Wispr @dreamy star
So the last Apple computer that I used was the Apple IIe
And I just dictated that to Wispr Flow and it formatted it perfectly. Wow go Wispr Flow! But anyways no, I use iOS for absolutely everything else so I've been thinking about getting a Mini Mac just for Clawdbot to dwell in.
But $20 for ChatGPT, $20 for Claude, $10 for minimax. All my subs are starting to add up!
Okay why are people paying for MiniMax?... This is literally the only server where I've ever heard anybody say that it's anything other than awful.
And I mean, facts are facts. It cannot play Snake. https://arc.markbarney.net/worm-arena/models?model=minimax%2Fminimax-m2.1
I’ve been using it as my main driver for clawdbot. And I use beads for project management on everything I’ve been give coding. Minimax is good for that too.
I’m just trying to conserve Codex and Claude limits for writing code.
Yeah I'm a project manager so that's like the meta game for me: managing my usage on $20 of Codex and $20 of Claude and then whatever I blow on OpenRouter or whatever playing Snake
I know people get amazing results using Opus 4.5 for everything but that's just not financially sustainable for most of us mere mortals 😄
Well here's the thing: if you have an OpenAI API key and you're kind enough to share your data with them, you get something like a million free tokens of GPT-5-Mini and GPT-5 Nano every day
I love clawdbot with opus. Personality is so much better. But stupid expense for most of the stuff I need a PA to do. It can just kick off codex or Claude for other stuff
There's an unbelievable amount you can get done with a million free tokens of GPT-5-Mini and it's not gonna be terrible
Oh wow. That’s probably a pretty good main driver
Is that one of those undocumented usage limits?
Maybe I’ll setup an openAI account just for clawdbot…
Once you have a PA setup with your skills and access I feel like you really don’t need a very large model
Yep I have to dig into it to find where the setting is, but I have like ten bucks on an API key for pretty much every LLM provider, just because some of them offer some kind of discounts or whatever from time to time
Like for a few months DeepSeek had this bizarre thing where tokens were like 40% cheaper between, I think, 6pm and 3am for my time zone
Odd. That’s like developer witching hours. Where all devs are hacking away
I haven't personally hooked it up to anything but Nemotron 3 seems very very capable, lightning fast, and free so I'm really not sure why people aren't experimenting with it nvidia/nemotron-3-nano-30b-a3b:free
I think the time zone made sense for China. I don't quite remember what it was exactly but I did find it funny
I haven’t used anything from Nvidia yet
Like I get a minimum 200 tokens per second with that model, the free tier on OpenRouter, which is ridiculous
Compare that with 2 TPS for GLM 4.7, between 12 and 20 TPS for things like DeepSeek and Minimax, and then 100 to 150 TPS for Gemini models
Have you tried OpenCode zen? Supposedly an open router competition
I mean we're all using stuff from Nvidia 😄
Ha yeah one way or another
Not heard of it. Is it like a Claude code clone, like Kilo code or RooCode or CodeBuff?
Yeah
But they are also starting to try to give out curated access to models. I guess some inference mechanisms on openeouter are not all setup the same
So I guess they try to give you better access to the same models you would get on openrouter but they test them someone
Seems like you would know more about that topic than me lol
I use AWS bedrock sometimes. And I swear their models seemed a little better than when I have occasionally used OpenRouter. Idk why. But probably sometime to do with what OpenCode is trying to do with Zen.
(I’m non-affiliated just an AI nerd)
I guess I've just used a lot of models last year 😄
Oh in what way? I mean I know that they have the exacto models, which are supposed to be tuned for tool calls. I'm kind of surprised people haven't been using GPT-OSS-120B with the exacto variety with Clawdbot.
nope i just found it one day on twitter haha. if it aint broke ...
Eh idk how to describe. I think I had originally used cline with opus via openeouter. Then I switched to Claude Code through bedrock. Could have been the hardness. 🤷♂️
I mean I'm kind of a normie. Like at some point last year around this time, I figured out, "Oh wait, I can hook this up to an IDE? I know what that is."
Like I know what VS code is and so thus making the leap to cursor and then to the cheaper version of cursor, Windsurf was not very difficult
Then it's just a whole rabbit hole of spinning up multiple terminals and multiple Claudes and multiple Codexes and multiple free models through Windsurf.
I was really big into cursor for a year. Now CLI is king for me.
For fifteen bucks a month on Windsurf you'll get 500 credits, which is whatever, but they give you an incredible amount of free access to models like GPT-5.1-Codex and stuff the pricing is literally insane.
Although I mostly still use cursor for work bc I think working in bigger projects requires a bit more planning that’s a little easier to do in a UI
Is windsurf still relevant? lol surprised they’re still around
Wonder how long anti gravity will last. I’ve heard they have some generous free tiers too
It's honestly everything that Cursor has and a little bit more, and way cheaper. I don't understand why more people don't use it
Like the code maps feature is free for paid plans so I make code maps of literally everything
That’s pretty cool