Hi there, trying to make it short:\u000a\u000a1. Have you heard about htps://jan.ai? In specific https://jan.ai/docs/local\u002Dapi allowing me to interact with it from any device (e.g. iOS shortcuts \u003D\u003E http request). I kinda read you are capable of that as well. \u003D\u003E True?\u000a\u000a2. I love “DRY” principles (Dont repeat Yourself). Currently I use a set of local makefiiles to manage my nested snippets to create dynamic prompts for the same Assistant (example: “Gandalf” could be act as the fully fletched assistant providing comprehensive answers, but sometimes I first want him to explain things to me like I am 5 without running the fully fledged model nor the fully fledged detailed respsone. I therefore got snippets for skills, formats, lenghts, …) \u002D sounds complicated but is easier to work with practically since I only talk to Gandalf and not 15 variations with different names for each assistant. \u000aShort: Can I nest (multiple) prompts of any depth as well as multiple snippet of any depth and ideally even get a “dynamic” prompt shown up initially with depending dropdown boyes allowing me to adjust any setting at runtime (and no, I don’t want to create 1001 keyboard shortcuts)\u000a\u000a3. Is there an internal API allowing me to get/post/push or trigger things (I read you mentioned webhooks but afaik webhooks only work on registered domains? Are those webhooks hosted on your end and if so whats the TTL (if I reboot and a webhook gets triggered while I am offline I want to be sure no request is lost but processed as I am back online)\u000a\u000a4. Since you mention local models: Those won’t require any credits since things run locally for those (as in jan.ai) \u003D\u003E right?\u000a\u000a5. Is your app context aware? Precise: If I select content on a browser website can I interact with it without copy/paste things first?\u000a\u000a6. Does it allow me to set “data sets” to refer to (e.g. a list of URLs or even a sitemap.xml) to perform actions on that or at least considering the content from those sites ? Ideally would be a mix of local files, remote files (on a gDrive \u003D not public content) as well as a list of URLs.\u000a\u000a7. A dream come true: For my articles and community I often need to create supporting drawings (svg based). Its about trading and explaining candle stick patterns, trends, divergences, etc.unfortuntately I dont own a full set of svg drwaings for each and every case and situation. While I found AI being usefull in explaining things, I was not possible to get any AI engine to be smart enough to learn from a set of svg drawings and adjust them. Is that that something I could use ALice for and train it on (create my custom model from a set of data she uses herself to learn from and getter on over time? If so, can I ensure a separation of concers (is she smart enough to understand in which context she works and not mix up the data set / models she adds “knowledge” to? \u000a\u000aI feel this is a bit edgy but thats what I deal with. I want my AI assitant to learn from things I flag as being relevant while knowing how to improve and not building up on things again and again that I marked as being wrong and iterated over it multiple times (\u003D learn from the iterations but remember the result so even if I start a new conversation, I don’t start from scratch / don’t require to load the full conversaiotn again and again (costly thing to do).

Question

Hi there, trying to make it short:\u000a\u000a1. Have you heard about htps://jan.ai? In specific https://jan.ai/docs/local\u002Dapi allowing me to interact with it from any device (e.g. iOS shortcuts \u003D\u003E  http request). I kinda read you are capable of that as well. \u003D\u003E True?\u000a\u000a2. I love “DRY” principles (Dont repeat Yourself). Currently I use a set of local makefiiles to  manage my nested snippets to create dynamic prompts for the same Assistant (example:  “Gandalf” could be act as the fully fletched assistant providing comprehensive answers, but sometimes I first want him to explain things to me like I am 5 without running the fully fledged model nor the fully fledged detailed respsone. I therefore got snippets for skills, formats, lenghts, …)  \u002D sounds complicated but is easier to work with practically since I only talk to Gandalf and not 15 variations with different names for each assistant. \u000aShort: Can I nest (multiple) prompts of any depth as well as multiple snippet of any depth and ideally even get a “dynamic” prompt shown up initially with depending dropdown boyes allowing me to adjust any setting at runtime (and no, I don’t want to create 1001 keyboard shortcuts)\u000a\u000a3. Is there an internal API allowing me to get/post/push or trigger things (I read you mentioned webhooks but afaik webhooks only work on registered domains? Are those webhooks hosted on your end and if so whats the TTL (if I reboot and a webhook gets triggered while I am offline I want to be sure no request is lost but processed as I am back online)\u000a\u000a4. Since you mention local models: Those won’t require any credits since things run locally for those (as in jan.ai) \u003D\u003E right?\u000a\u000a5. Is your app context aware? Precise: If I select content on a browser website can I interact with it without copy/paste things first?\u000a\u000a6. Does it allow me to set “data sets” to refer to (e.g. a list of URLs or even a sitemap.xml) to perform actions on that or at least considering the content from those sites ? Ideally would be a mix of local files, remote files (on a gDrive \u003D not public content) as well as a list of URLs.\u000a\u000a7. A dream come true: For my articles and community I often need to create supporting drawings (svg based). Its about trading and explaining candle stick patterns, trends, divergences, etc.unfortuntately I dont own a full set of svg drwaings for each and every case and situation. While I found AI being usefull in explaining things, I was not possible to get any AI engine to be smart enough to learn from a set of svg drawings and adjust them. Is that that something I could use ALice for and train it on (create my custom model from a set of data she uses herself to learn from and getter on over time? If so, can I ensure a separation of concers (is she smart enough to understand in which context she works and not mix up the data set / models she adds “knowledge” to? \u000a\u000aI feel this is a bit edgy but thats what I deal with. I want my AI assitant to learn from things I flag as being relevant while knowing how to improve and not building up on things again and again that I marked as being wrong and iterated over it multiple times (\u003D learn from the iterations but remember the result so even if I start a new conversation, I don’t start from scratch / don’t require to load the full conversaiotn again and again (costly thing to do).

overment · Answer

\u003E Have you heard about htps://jan.ai? \u000a\u000aYes, and I\u0027ve heard a lot of good things about this.\u000a\u000a\u003E  In specific https://jan.ai/docs/local\u002Dapi allowing me to interact with it from any device (e.g. iOS shortcuts \u003D\u003E http request). I kinda read you are capable of that as well. I kinda read you are capable of that as well. \u003D\u003E True?\u000a\u000aNot exactly, at least not now. Let me explain.\u000a\u000a\u002D Jan.ai provides a built\u002Din local server that listens for INCOMING requests. It\u0027s like sending messages to Jan.ai.\u000a\u002D Alice lets you CONNECT to external automations through webhooks in Remote Snippets and receive a response. Alice\u0027s \u0022Custom AI\u0022 feature enables you to connect to your own server, including localhost. This turns Alice into a visual interface for your custom solution, offering powerful capabilities.\u000a\u000aTo put this simple: Jan.ai can listen for external requests, while Alice doesn\u0027t. However, you can interact with external services and your own server through Alice\u0027s interface.\u000a\u000aI hope that\u0027s clear. It was challenging to explain because the difference is subtle and technical.

overment · Answer

I accidentally sent a comment, so I\u0027m continuing here. I\u0027m limited by character count.\u000a\u000a\u003E Can I nest (multiple) prompts of any depth (...)\u000a\u000aYes and no. You can set the Assistant\u0027s general prompt and assign Snippets and Remote Snippets to extend its behavior. You can change active Snippets during the chat. I use this option in a similar way to what you\u0027ve described.

overment · Answer

\u003E Is there an internal API allowing me to get/post/push or trigger things (I read you mentioned webhooks\u000a\u000aCustomAI may serve as an internal API, but it\u0027s a feature for programmers. Here\u0027s a template to start with: https://github.com/iceener/heyalice\u002Dnodejs\u002Dbackend\u002Dtemplate\u000a\u000a\u003E but afaik webhooks only work on registered domains?\u000a\u000aNope, it may be localhost or webhook generated with make.com for ex.

overment · Answer

\u003E Since you mention local models: Those won’t require any credits since things run locally for those (as in jan.ai) \u003D\u003E right?\u000a\u000aYep. CustomAI works the same way unless you connect your server to a paid service like Anthropic or OpenAI.

overment · Answer

\u003E  Is your app context aware? Precise: If I select content on a browser website can I interact with it without copy/paste things first?\u000a\u000aNope, you need to copy and paste it first. To make it easier, you can assign a Snippet or Remote Snippet to a global keyboard shortcut, making it simpler to use. It\u0027s super handy, and I use it all the time (even right now).

overment · Answer

\u003E Does it allow me to set “data sets” to refer to (e.g. a list of URLs or even a sitemap.xml) to perform actions on that or at least considering the content from those sites\u000a\u000aOnly through Custom AI, as creating a generic, ready\u002Dto\u002Duse integration for all users is beyond my reach. You can create your own local server to handle tasks and use Alice as the interface for interacting with it.

overment · Answer

\u003E While I found AI being usefull in explaining things, I was not possible to get any AI engine to be smart enough to learn from a set of svg drawings and adjust them (...)\u000a\u000aI don\u0027t know if it\u0027s possible with the current generation of LLMs. I\u0027ve created some custom graphics using ChatGPT Code Interpreter or Claude Artifacts, but they\u0027re not quite the SVG drawing I envisioned.

overment · Answer

\u003E I feel this is a bit edgy but thats what I deal with. I want my AI assitant to learn from things (...)\u000a\u000aIt\u0027s outside Alice\u0027s current scope, but I have my own, private \u0022Custom AI\u0022 server that does most of the things you\u0027ve mentioned. It has short\u002Dterm and long\u002Dterm memory, a set of tools, and can \u0022reason\u0022 about things.\u000a\u000aIt\u0027s called an \u0022agentic system.\u0022 You can read more about it.

nl8 · Answer

Thanks for getting back \u0026 taking the time to respond. I like the option to trigger external automations \u0026 receive their response but seriously would love being able to e.g kick off things while browsing without switching context (\u003Dsend it to my assistant rather switching wnidows. To name a max simplified case). Getting Alice to “listen” would be seriously maximiizing her value even more.

Alice

Share Alice

Related questions