What are the best AI options right now?

Questions
22

yeah my boss and I have been testing it out via openrouter.ai and have had a good experience.

1 Like
23

Curious how this one compares to the following: https://opencode.ai/

1 Like
24

oh cool, I am excited to check it out.

1 Like
25

It seemed like a cool alternative until I realized Brave had blocked over 99 trackers. I got rid of the account fast.

28

As far as I know, you can put any API key for your proprietary paid models. :slight_smile:
Would be totally pointless otherwise. :sweat_smile:

1 Like
29

I don’t know what I was thinking..

1 Like
30

I agree that’s a bummer, at the same time they offer legal guarantees that your data wouldn’t be used with the Zero-Data Retention.

I think Gemini models are the best quality/price ratio. Chinese model Kimi 2.5 is also quite good.

Don’t restrict yourself to one models. Use multiple models for important tasks.

1 Like
31

What about offline/local AI apps? Which are the best and what are you guys using?
For Linux, I’ve read about Newelle and Alpaca, both available as flatpaks and looking good. I’m in the market for some offline apps on mobile. Especially on iOS. I’ve only tested Ensu from Ente. What would be the best options there and its Android counterparts?

33

Has anyone tried okara ai yet? Private ai using lots of open source models.

34

I personally use Brave Ask, Lumo and Perplexity (in incognito mode) in that order. I never include any PII and Perplexity is a last resort or for when I need properly thorough research. I usually find Brave to be more than sufficient and 90% of queries go through that. I’ve dabbled with self hosted solutions but I find them slow and a fairly poor experience. I think the privacy benefits are outweighed by that experience.

35

To run LLMs locally you can use ollama on Linux and PocketPal AI on Android. Ente’s Ensu is also an interesting development, but currently you can’t choose which LLM to run.

36

Edge Gallery is good for android. But I don’t use local models anymore. They are too slow honestly.

It looks good overall, but a few thing I would like to see

  • Founders name, more company info. Company is based in Singapore
  • Opt-in to not sync chat data in their servers. That’s what OpenRouter does and it’s better that way. Keeping message local is better than encrypted in the cloud.

Also, the model is completely different from OpenRouter. Here, you send all your prompts and data to the company, hoping they keep their promise. OpenRouter routes you to many AI providers, and those with zero-data retention if you choose.

37

Finally got around to giving this a go. Really like it. Even just using my openrouter API key, its a much better interface to use it from.

38

I use https://chat.z.ai. It is unlimited, free, no sign up, about as good as latest gemini, claude in my experience, way better than chatgpt. It is chinese, so no discussion critical of China.

I wonder if there is any way it is a good tool for people who wouldn’t care if china got their data, but didn’t want the US to have it. Although I could see them needing an agreement with the U.S. to be able to run it here. Curious if anyone else has thoughts on this.

1 Like
39

Z.ai has most of their model in the open, including GLM-5. I recommend using their models with third-party tools instead.

40

Do you have a specific tool in mind you recommend? I mostly just use chat.

41

Open Router, as mentionned above.

I guess you could also just chose a provider (see a list on openrouter.ai/models, filter for Zero-Data Retention) and see if they have a web offering.

42

Really, no one mentioned venice.ai ??

Venice PWA for LLM-things, Brave Leo for browser stuff

Edit: and if you have to use it for work, then they can pay for Pro sub

43

I’m currently splitting between:

  1. small local models (my hardware is insufficient for more than that) [most private]
  2. nano-gpt (restricting myself to only using the TEE models) [more private]
  3. Duck[.]ai (for queries that aren’t very sensitive or private) [somewhat private, with caveats, largely trust based]

They basically act as a proxy to whatever ai service you call. And while maybe that protects you directly, it doesn’t prevent the query itself from being logged by the AI.

Yeah, I think that’s mostly accurate. But if you think about it, that’s essentially the same as private search as well.

If you can’t run it locally, Nano-GPT is an option.

I don’t recommend it (nor do I recommend against it) but I do personally use it.

I use it because I can (1) pay anonymously (2) they offer TEE[1] models, (3) pay-per-token pricing option (4) can be used via the website or API

  1. Trusted Execution Environment / Secure Enclave / Confidential Compute ↩︎

2 Likes
44

Why not Proton Lumo? Any specific reason for choosing Duck?