WikiSage -- Your Tiddlywiki AI Companion

As hinted in the topic for the Expanded Chat GPT Plugin, I have been toying around with some features that have called for changing the naming schema - - So, here I want to announce the pre-release of WikiSage – Your Tiddlywiki AI companion for content creation, organization and interaction

At the link above you can see my first github page, or you can download the plugin directly:

WikiSage.tid (1.3 MB)

One can expect all the same features from WikiSage as can be demonstrated at the Expanded Chat GPT Plugin page, but with some improvements as well as those name-breaking features:

As you can see, this update allows the user to query Anthropic models in addition to the openAI models that the original plugin did. It handles the switch between API calls elegantly so the user doesn’t need to think about it and can switch back and forth between openAI and Anthropic models using the dropdown menu.

Eagle-eyed readers will also have noticed an additional button in the container:
image

This button allows the user to upload PDFs for review by the agent. Please note that this feature will actually require the use of an Anthropic API call regardless of which model one is choosing to converse with - this is because PDF review is a beta feature of anthropic which is only available currently for 3.5 Sonnet and not available at all with OpenAI (yet).

The plugin handles this by automatically switching to 3.5-Sonnet if it detects that the user has uploaded a pdf, and then automatically switches back to whatever model was previously being used.

All other features, including TTS and STT are compatible with the Anthropic models, except for Export, which is currently still being retooled for Anthropic.


Excited about the direction that this new integration is taking the project and though I have several improvements and new features in mind, I’m very interested to hear from others about their thoughts :slight_smile:

There is still some work to do before a “version 1” release – other than fixing the export tool, I also need to go through and change all the naming schemas in the widget to match the new titling (you will note that the widget is still called using the same name as the previous version), then thoroughly test that system to make sure I haven’t broken anything…

In addition, I would really like to wait until after the release of 5.3.6 to do the bulk of that testing, so currently this should be considered a pre-release with the latest version of Expanded Chat GPT Plugin considered stable.

But, frankly, I couldn’t wait to share :slight_smile:


Also, please note, this is a major repackage and also my first time posting to GitHub. Though I have tested some, it is entirely possible that I have missed something – if you encounter an error, please let me know and I will resolve it in a timely manner.

4 Likes

I have something to share, too. WikiSage is a really cool name :sunglasses:

1 Like

@linonetwo, I know you were asking about Claude integration in the AI Tools plugin and @jeremyruston noted that it was difficult to find documentation on http API – in building this out, I discovered that this is possibly because Anthropic did not even allow CORS until late this year.

Have not looked much into Gemini’s system yet, and I’m not particularly eager – at the moment, ollama is far more likely to be the direction I’m looking.

I appreciate that, I sweated over it a little :sweat_smile:

Thought about calling it “ephēmeris” but figured that might be a little obscure…

Hi @well-noted could you implement the api of deepseek
https://chat.deepseek.com/ It is cheaper than claude or chatGPT and pay-as-you go and for some issues it is perfect.

And great you integrated https://claude.ai/ as well. When I am thinking on a suscription it would be rather claude than chatGPT at the moment. It performed much better with the coding tasks I had. And it seems to have a more ethical policy.

Hey @JanJo , I have heard a little about deepseek recently but have not delved too deeply. I can see they have an API, so if they allow for the same sort of CORS policy as openAI and Anthropic, then it would likely be possible.

However, I will remind you that both anthropic and openAI models are “pay-as-you-go” in the sense that each query only uses tokens respective to the length of the query/response. You pay upfront for, say, 5 dollars worth of tokens, and then you can query freely until that amount of tokens is expended. There is no subscription required though, the subscription for either is entirely separate from the API calls.

I still have not gone through the 20 dollars I put on my account prior to starting the project.

Thank you- I felt so hussled with the subscription offers, that I did not look for the API
But it would be cool anyhow to have deepseek as a possibilty.
And as a german teacher I would love to have the IONOS Api implemented because this is DGSVO which meas subject to the General Data Protection Regulation which is necessary to be used in places like schools in europe.
I bet deepseek is not

1 Like

Hi, @well-noted IONOS claims to be compatible with OpenAI
And not to forget: The next month is still free of any cost.

I would be especially hesitant to be sending personal data onto Chinese servers… And I’m not seeing very much about Deepseek to make it especially appealing over the existing options

@JanJo, it does seem like Hugging face has an open source version of the model that can be installed and run locally.

A good reason to get the oLlama version working.

1 Like

I think the release of Deepseeks R1 as OpenSource is one of the reasons for the big buzz.
But I cannot find clear instructions what ressources it needs - so far I got no 3090GPU.
As said for privacy reasons I would even be more interested in unsing the IONOS-cloud.

Hi @well-noted how do I get wikisage going with anthropic. I entered my Anthropic API-Key, and it still asks me about my openAI AI (that I do not have yet).

I see this, @JanJo, and have the space today that I can provide a walkthrough :slightly_smiling_face:

1 Like

@JanJo, before I get too wordy, will you please try something for me?

Locate or create the shadow tiddler titled $:/plugins/Notestreams/Expanded-ChatGPT/model-list

in that, please paste:

gpt-4o-min
gpt-4o
gpt-4
gpt-3.5-turbo
gpt-4o-2024-08-06
dall-e-3
o1-preview
claude-3-opus-20240229
claude-3-sonnet-20240229
claude-3-haiku-20240307
claude-3-5-haiku-latest
claude-3-5-sonnet-latest

After doing that, you should see that list reflected in the model list dropdown

If you don’t, let me know. It is possible that I have an update I need to push, but you should be able to add any openAI or anthropic model to that list, and it will route the api correctly.

You should also be able to add the list and any additional models to the model list from the config tab.


It seems the edit-box for the model list, within the config tab of the plugin is still using the ExpandedChatGPT reference, rather than the Wikisage reference. I’ll need to dust off my working version and push an update.

That’s probably timely, in any case :slight_smile:

I just pasted the list, now the claudes appear in the dropdown - but I do not get clear visual feedback what I have chosen and an alert still asks for the openAI api

Please send a screenshot? :slight_smile:

This is what I see when I select the o1 model and then click the dropdown button again:

image

This indicates the model that is currently selected.

You are correct that this is not the most communicative feedback, I am thinking I will incorporate a small indicator next to the chat about which model is selected

Well it seems that I chose haiku… but it still asks for the openAI API-Key

Hmm. Can I see that openAI Message?

Please add anything to the openAI edit box, which I imagine should make the message go away.


Have confirmed that this is how it works on my empty. I see the message, it is an artifact from when it was only OpenAI, it will be resolved in the V1 :slight_smile:

1 Like

I am afraid I can’t, the alert crashes the screenshot addons.


It really cures the problem to input anything.

1 Like

I appreciate you taking the time to talk this through with me, there haven’t been many people who’ve tried it, and apparently you’re the first to attempt to use Claude without first entering an OpenAI key :slight_smile:

I’m adding it to my list of V.1 fixes

1 Like