AI Assistant: Generative AI / GPT3 Integration
complete
N
Nicolo S
Integration with OpenAI's GPT3 model (for paying users) to allow users to highlight a sentence or paragraph and leverage OpenAI to e.g. rephrase, summarise etc.
Log In
Steffen Bleher
complete
A beta version of the AI assistant is live. 🚀
You can read about it here: https://capacities.io/whats-new/release-26
M
Marc Volquardsen
Steffen Bleher: After you have given us the different email-addresses for each of the spaces it will be the same request here.
Different OpenAI API Keys for each of the spaces. At my company I have a payed account for OpenAI and I want to use this. But I'm not allowed to mix the data with information from my private space. There I would work with your API Key.
And another thing. In business we rely more on the Azure OpenAI Services (in Frankfurt).
We also have opt out, so Azure doesn't forward any data to OpenAI for training of their system. So providing an other endpoint than the standard OpenAI endpoint would be very helpful.
Steffen Bleher
Marc Volquardsen: Thanks for sharing. This is notes. We'll think about it :)
F
Fer Girante
- Self organization like Mem.ai and Napkin, should be a must. We spend way to much time organizing notes, links and tags
- Being able to "talk" with your notes and "ask questions" to your notes would be a killer feature
Regards
v
vision assemblies
Fer Girante: hm. I also value Napkin for what it does. but the thing is: it works quite differently at
its very base
, meaning this kind of "self organization" is actually it´s core proposition – coming at the – very relevant cost (trade-off) of it being occlussive (as to its relational structure).Capacities on the other hand has a quite different core proposition and basic metaphor to deliver on (think: Studio for the Mind). here things live of visibility (and different viewing lenses
to be chosen
), controllable
spatial organization, selectable
transclusion etc. ...Still, automated semantic things like "related objects" (which has just been marked as "planned" – as "smart link suggestions / unlinked mentions") is useful indeed. – see: https://capacities.io/feedback/p/smart-link-suggestions-unlinked-mentions
but things should always be appropriate to the context / architecture of the app in question (Capacities). so, I would say, it all depends on, what part / kind of instantiation of such a big term as "self organization" you deem as being "a must".
[btw: as to "We spend way to much time organizing notes, links and tags" – you might want to see yourself confirmed in a nice way here: https://www.youtube.com/watch?v=0YkiGw7G0d0 (– or a discussion of that in Capacities context sharing your main sentiment: https://discord.com/channels/940596022344843336/1001365576314261555/1092809291850842194)
as to the now proven advantages of being able to have LLM-AI integrated with minimal interface like a chat / command box, I also agree. (though I always tend to disagree, if someone talks 'killer feature', as it mainly is an index for things one person really badly wants :-D)
N
Nerdy Momo Cat
Personally, I would like to see a mixture of mem (write in your voice based on the information you have in documents), tana (auto-fill and auto-link objects and mentions based on fuzzy matches and break them down into components) and heptabase AI (ask questions and get a search page of responses from information in capacities) implementations.
The thing I have 0 inclination towards is Reflect like AI integration which is basically copy pasting a prompt and text and getting the answer to output into the document.
Steffen Bleher
Hi everyone,
we are currently collecting ideas and thoughts on the AI assistant. You can share them here or in our discussion on Discord: https://discord.com/channels/940596022344843336/1120340681102204948/1120340681102204948
Super excited to get this started 🚀
Hi Steffen,
In assumption that capacitites uses their own (OpenAI) API key and aggregates (and averages out) the total cost of these API calls in a general premium memberships account, I would like to suggest the following to have a future proof system:
- Can you also allow users to use their own OpenAI key to use other models where required (as in GPT-4 or other simpler versions of their models)?
- Can you allow selection of which model is used in the settings?
- Can you enable a plugin system where API keys, API endpoint is provided plus some scripting to allow integration of alternative AI-tools? perhaps people would be interested to use Text-To-Speech, or image generation or anything else in the future!
Steffen Bleher
in progress
D
David Cu
I’d like to see a “BYOK” option so I can integrate my own OpenAI key and settings. I find that everyone is integrating some sort of AI but without any customization options in how the LLM behaves.
Steffen Bleher
David Cu: Thanks for sharing. What other customizations except from the model itself are you thinking of?
M
Mike Clarke
Steffen Bleher: Hey, not OP but wanted to also get my feedback in on this. We definitely need an option for our own api key, that way we can use GPT4 or other models (other wise capacities monthly cost would skyrocket?). The best implementation of AI I've seen so far is Coda, they do a lot of smart stuff with gpt. But even just the basic generating text within the content blocks on page, select text, have default prompts show up, and option to just type what you want (custom prompting). I'd also really like to see context feeding, like using @aPage within a prompt to send the content in that [page] to GPT along with the prompt, that would be super powerful combined with GPT3.5 16k on large documents. Ability to save custom personas and select which to use before sending the prompt would also be great, just a basic text field that is sent as a system prompt before the users prompt.
Basically a text gen playground :D Love Capacites thanks for all the great work, looking forward to seeing this become the PKM powerhouse.
Edit: Also if you hadn't already noticed I'm super keen, so if you need a alpha tester that will provide great feedback I'm your man 😁
S
Syahrenzy Olov
Answer questions or give summary based on all notes (with/without filter) when prompting through Search & Command Palette. Can also continue the chat through sidepanel
V
Victor Ngo
Making, saving, and editing custom prompts would be awesome as well.
Steffen Bleher
planned
Load More
→