# Welcome to Riku.AI Riku is the vault of your A.I. creations and a place to save, distribute and experiment with the latest technology. Our team got tired of switching through multiple tabs and pages so built Riku. You may be wondering what Riku is all about. There are a ton of different products coming to market which effectively wrap the API of a technology provider, give some half-assed prompt and let you use their generator. With all of this technology becoming a lot more accessible, the real value is in the data, the prompts and the minds behind those who can figure it all out to create greatness for their creations. We're all about building a place to help you succeed. Riku is an aggregation of all the best tech on the market enabling you to play seamlessly with it in one easy-to-use playground. Switch the tech behind your models to see what performs best and which gives the best bang for your buck. Riku is a blank canvass for you to build upon. It provides you the tools to create and share with the community and we're so excited to see what is built! {% hint style="info" %} Riku allows you to generate text from OpenAI, EleutherAI's GPT-J, Cohere and AI21 in one single playground without any filters. {% endhint %} # Connecting Keys All Riku accounts come with GPT-J connected using our infrastructure, for the other providers, you will need to add your own keys. Adding your own keys does not need to be difficult. We will provide a quick walkthrough on how to get the keys from each of OpenAI, AI21 and Cohere in the next few pages. To add these to Riku, you will need to make sure you are on the [API settings](https://riku-app.bubbleapps.io/version-test/dashboard/api-settings) page. {% hint style="info" %} Security is super important to us. Your keys are kept encrypted and separated from anything which specifies what they are used for. You are also able to delete your data at anytime, we do not keep a record of it once a deletion is made. {% endhint %} From here, you will see three boxes where you can add your keys. You won't be able to use these technologies on Riku until they have been added so we'd advise you to add them, if you do not then you can always keep using GPT-J. # Connecting Aleph Alpha Keys Getting setup with Aleph Alpha is very simple. You can go to their website [here](https://app.aleph-alpha.com/) to sign up. Once you have signed up, you will be given some free credits to play around with. You can also then get your key. You can click on your profile in the top right of the page, or if you are logged in, click [here](https://app.aleph-alpha.com/profile) and from this page you can see your token. Click the "Copy" button to copy it to the clipboard and go back to Riku and your API settings page, or click this [link](https://riku.ai/dashboard/api-settings). Paste the code in the box for Aleph Alpha and hit save / update. You will see a success message and be all set up for using Aleph Alpha within Riku! # Connecting Muse API Keys First of all you will need to sign up for a plan at . You will be required to enter card details for this but the free plan is extremely generous. You can see our first look on the pricing and website here - . Once your account is created and you are logged in on the Muse site, you will see "Account" in the top right. Click on this and choose the option for "API Keys". ![Create a new key and name it Riku. It is that simple!](/files/eOuo5qBmRQcVJ3rOinCw) From here, you can hit "Add key" in the top right and you will be able to give your new key a name. Call it Riku or whatever you want and you will then be able to copy it to the clipboard. From here, go back to the Riku Dashboard and the API Settings and enter it in the Muse API box and hit save. Your connection will now go green and Muse is ready to use within Riku! # Connecting OpenAI Keys OpenAI was invite only for so long but that has all changed making connecting your OpenAI key to Riku super simple! OpenAI keys allow you to use the multiple models that OpenAI offers. If you do not have an OpenAI account, you will need to create one. You can do so by clicking [here](https://openai.com/api/) and hitting signup. Follow the instructions to make your account. Once you have created your account, you will be able to get your key from your settings page. ![](/files/2fXhQJcgvYtvc7Ti4Biz) If you are logged in, click this [link](https://beta.openai.com/account/api-keys) directly to get taken to the page with your API keys on it. Simply click the copy button and then paste that into the OpenAI settings [here](https://riku.ai/version-test/dashboard/api-settings). It can be good practice to create a new secret key for each service you link up so if you want to do that, hit the button to create a new key in OpenAI and then copy across that new key. You can delete your key at anytime from OpenAI which will break the link with Riku, also if you are so inclined you can remove the connection directly from your Riku account. # Connecting Cohere Keys Cohere follows a similar setup to OpenAI and requires an account for you to be able to use the key within Riku. The easiest way to do this is to head over to the [Cohere website](https://cohere.ai) and hit the Get started button in the top right. Direct link is [here](https://os.cohere.ai/register). Once you have an account, you'll be in their old school operating system design. If you look at the bottom box on the screen, you will see API Keys. You can create new ones and give them a title. Why not create one called Riku? ![](/files/Ew0SXoLNyAwKe0JYaKSn) After creating this key, you will be able to copy it and go back to your [Riku dashboard](https://riku.ai/dashboard/api-settings) where you can enter it in the settings. This will now enable you to use all of the Cohere models within Riku. # Connecting AI21 Keys For AI21, you will need to create a Studio account. You can do so by signing up [here](https://studio.ai21.com/sign-up). Once signed up you will be taken into their own playground. You will want to click the user icon in the top right and click on the Account menu that opens up. Alternatively, click this [link](https://studio.ai21.com/account) directly. ![](/files/kBmuxtoQ9FtewW3zmNO4) At the top of this page, you will see your email address and your API Key beneath it. You can hit the icon to copy your key to clipboard and that is as simple as things get! Then go back to Riku's API settings and paste it in the AI21 box. You can find that page [here](https://riku.ai/dashboard/api-settings). # Defining Key Terms Throughout this support guide, I will refer to specific terms which have a place in AI and technology as a whole. If you see a term that you do not know the meaning of, you may want to come back to this section to look for the definition. I want to make this e-book as simple and user-friendly as possible using common language where possible. **Prompt** - When creating an AI model, you need to tell it what you want it to do. This is the prompt. Think of the prompt as instructions. An example could be ‘Write me an introduction for a blog post based on the following information’. **Preset Models** - This is where you are accessing an AI model via a third party for example Content Villain. You are using a model that has been pretrained based on the determination of the company you are using it through. **Model** - Different companies offer different models for generating text. These can vary based on size (parameters) and also on the quality of the training data. Generally all models are in the billions of parameters size range. Size doesn’t necessarily mean best quality for all tasks however so digging a little deeper can be valuable. **Prompt Engineering** - This term references where an AI model is created by providing a few examples within the prompt. This can also be referenced as ‘few shot learning’. You are expecting **Finetuning** - The method of finetuning is when you provide an AI model with a large dataset of examples. By providing such a dataset, you are giving the AI a better understanding of what you are wanting it to output and can remove the need to prompt engineer. Generally speaking, finetuning will get you a better quality output if your dataset is of a good quality. **Datasets** - Datasets are large files of text with multiple examples of your desired AI output and also the input text. An example dataset could look like the following; if you wanted a dataset for blog introductions, you could provide 50,000 examples of a blog title and brief description and also 50,000 examples of the blog introduction. **Tokens** - Most companies offering AI Models work on tokens for pricing. Each token is similar to a syllable so a longer word may take multiple tokens to create. It is also important to understand that some companies include input tokens alongside output tokens in their pricing. # Winning with Text Based AI Creating text with AI will speed up your content creation, make your SEO better and ultimately make your business life easier. It can make a lot of difference but is not without some limits. We want you to have the most success with AI and creating content for your business and will be here to help you with the information you need for that success. * **Keep it Simple** - Don't try to create too complex content at one time. Train the AI to do one thing well and think about chaining multiple outputs together for longer content. * **Longer Content is Unpredictable** - The higher the output tokens you set, the more unpredictable that output will become. The AI works by trying to predict the next token in a sequence, the more freedom you give it, the more chance of it turning in a wild direction. * **Patterns are Everything** - Teach the AI a pattern and it will follow it forever. Reinforce that pattern with some examples and it will make the output better quality. Breaking a pattern will lead the AI to give you a poor quality output. * **Use Quality Examples** - Examples are what the AI will base its outputs on. There is no golden bullet in terms of number of examples to provide but the AI will want a wide array of diverse examples to pull from for the future outputs. # Prompt Building 101 Prompts are everything. With a prompt you are telling the AI what to do, and you are also providing it examples of what you expect it to output. There are a few tricks to understand when building prompts. Here we will go over some of the most important. 1. When setting up a new prompt. Think very clearly about what you want the AI to produce. Be descriptive and to the point on this. If you want it to create a product description for a clothing brand, you might want to write 'Write a product description for a clothing brand' as the first line for your prompt but you can improve this massively by going into a bit more detail. This would get you better outputs; 'Write an exciting product description for a clothing ecommerce brand which focuses on women's clothing for the ages 18-30'. The added detail will make a better output. 2. Think about your inputs. Inputs are important as it gives the AI the context of what is needed when producing the output. The AI will produce some magical content but it isn't a mind reader. It needs information to base the output off. If we carry on with the same theme as above, you could just have the item title in your prompt like Title: Frilly Black Dress and hit generate. The AI will write something but it won't know enough details for any accuracy so you may want to add a second input with descriptive details where you can put information on the length, the fit, the style etc which gives the AI better background to work with. 3. Keep the pattern with the stop sequences. Generally, most people building prompts with AI will be familiar with the default of ### which is not a bad thing to use. Between each set of inputs, put a ### and also between the outputs. You could really use anything for your stop sequences but following this pattern will help to standardize your creations and make your prompt building faster. 4. Examples are key to everything. After your initial instructions at the top of your prompt and you have then decided on your inputs, you can then build out some examples. This can very much impact the quality of what comes out of the AI. We'd recommend anywhere from 2-10 examples based on the length of the output desired and the type of content you are trying to create. 5. Set your maximum tokens correctly. The easiest way to do this is to do a character count on the examples you have given in your example. Take the example with the highest character count and then divide this by 4. That is the value that you want for your token amount. Based on if it is a large number or not, you might want to add on 10 or 20 just for any issues but that will be roughly correct. # Common Issues & Errors Creating prompts with AI definitely has a knack to it and you may not find it super simple right away. It is however very rewarding and in a way, similar to writing code. Any single mistake in your prompt could cause an error and these are what we want to avoid. Here we want to provide some of the common issues we encounter and how we should avoid these. * **Maximum Prompts** - Every model provider has a maximum prompt output. When counting the output, they take the prompt into account, this means that you may have 2048 tokens to play with. If your prompt is 1500 tokens and you are trying to generate 700 tokens, you will be over that 2048 total so it is a balancing act of proviidng enough context within the prompt but also allowing enough tokens to be used for a quality output. {% hint style="info" %} A token is roughly 4 characters. What we like to do is to take our prompts and put them into a character counter. Divide the number by 4 and you will have a rough estimate for the token amount. This won't be perfect but will be near enough for you to use when calculating your models and ensuring you don't go over the maximum. {% endhint %} * **Lack of Context** - It is easy to think of this technology as groundbreaking and to big it up more than it is. You might make the big brain move of thinking let's just make it write us a full blog and putting in a prompt like 'Write me a full blog on Financial Systems' and then setting the token amount to about 10 less than maximum to take into account that title and hitting generate. The AI will generate those tokens but it has no context and limited idea for what you want so don't expect anything useable to come out. Examples are often required and encouraged for quality prompt building. * **Filters** - Due to the way we have build Riku, there are no filters on the majority of our models. Note that if you are to go to production with anything from OpenAI, you will be required to put a filter on the model but in RIku, we don't use them as we just provide the playground for building out your prompts. Cohere is the only technology that has the filter built directly into the generate endpoint so you may encounter filters with Cohere whilst using it in Riku. * **API Keys** - Check that your API Keys are correct when entering them into Riku. If they are not added correctly, the technology will not work. You can remove the current settings and add your new settings at any time. # Ethical Considerations This is a new technology and there are a lot of ethical considerations. Think about the content you are trying to create and what it means for the people who will be interacting with it. A lot of the content made online is made for the purpose of SEO and attracting the approval of a Google robot. This is not going to change. It is worth however considering any content that will directly interact with a human. Should they know that the content they received is not generated by a human? How far do we go? Politics, fake news, explicit content, violence, drugs, where is the line to be drawn? When signing up for some of these commercial providers, you will see terms and conditions for the content that can be created. OpenAI and Cohere have quite strict content filters on anything explicit and violent. Others do not want their technology used for longer term content creation. What you create is ultimately down to you but please consider the ethical arguments when playing with this technology and make a decision that works for you. That is how we like to play with this technology anyway. # Single Endpoint Export Anything that you create within Riku or that is available in the community showcase can be used outside of Riku in your own projects or workflows. Why should this excite you? Exporting prompts enables you to have full control over your AI creations. You can use them in your own applications or workflows. You are not restricted by what is available in Riku and you can build out more complex flows to suit your needs or instantly get the outputs where you need them. It might seem daunting if you are not a coder as to how to do this but it is really quite simple and we try to explain it as easy as possible so you can set it up for yourself whether you are a no-code professional or have more extensive knowledge of code. {% hint style="info" %} One of the common tasks that we get asked about is how you could set up a process where you enter some inputs and get the AI to output directly to a spreadsheet or Google Sheet. This is very simple to perform and we'll give you all the tools you need to succeed in setting up this process. {% endhint %} Having the ability to use your prompts on your own terms is incredibly powerful. There are a few different steps that we have to go through to unlock the full potential of using your prompts where you need them so let's get started! # Getting Prompt IDs The first step of using a prompt in your own workflows or processes is finding the prompt ID. This is very simple to do! You can use prompts that you or your team have created and you can also use prompts from the community showcase. Click on the one you wish to find the Prompt ID for and you will see on the left side a purple button for Copy Prompt ID. Click this button and get the success alert. ![](/files/yerWwVPechggUurZm1kp) Awesome! The Prompt ID is now copied to the clipboard and you can use it in your own flows. The next thing we need to do is to know the structure of how to use this. {% hint style="info" %} Whilst you are on this page copying the Prompt ID, it is important to also note the input fields and what they are expecting. You can see in the example above, this prompt has no inputs which will make things easy as we go forward and now build out the API request. {% endhint %} # Structuring the Request Now you have the Prompt ID, you can build out the request to use in your own system. We like to test all of our API requests using a tool called [Postman](https://www.postman.com/). Well worth picking it up if you don't have it. It is free to use. What Postman is great at is that it lets you import a CURL request and then try it out instantly. This is great for experimenting and ensuring you have everything set up properly. You can then also get a code snippet for multiple other technologies such as Python, Javascript, NodeJS, and much more. ``` curl --location --request POST 'https://prompts.riku.ai/webhook/run' \ --header 'Content-Type: application/json' \ --data-raw '{ "Name": "NAME OF ACCOUNTHOLDER", "Secret": "SECRET KEY (GET FROM MANAGE ACCOUNT)", "Prompt ID": "COPY FROM RUN PROMPT PAGES", "Input 1": "THE FIRST INPUT", "Input 2": "THE SECOND INPUT", "Input 3": "THE THIRD INPUT", "Input 4": "THE FOURTH INPUT", "Input 5": "THE FIFTH INPUT", "n": }' ``` This is the request you can use for anything you want to use from Riku. The POST request is going to an endpoint regardless of which technology the prompt is using and we will route it accordingly for you. {% hint style="info" %} In the code example above, we provide data for Input 1 - 5. Our system will automatically exclude any Inputs that are not relevant to the Prompt ID specified. If you have a Prompt ID with zero inputs, you can still include the Input fields but they won't be used for anything. {% endhint %} The main points to keep in mind for beginners when using this request are; * It is a POST request. * There is one header field of "Content-Type": "application/json". * The "n" field is a number and generally, we don't recommend putting that as more than 6. Also, ensure that you do not put that number in quotations. It should be "n": 5 and not "n": "5". Final note to remember is that the final field of the JSON does not have a comma at the end. # Understanding Input Fields Now we have seen the structure of the export and we know how to get the Prompt ID using this request in your own workflows or processes is super simple. You may not understand what all of the fields are so this section aims to break them down into simple terms for you. * **"Name"** - When you setup your Riku account, you could specify your name. If you have created any prompts and added them to the community showcase, you will see what your name is set as here. This is a required field and you won't be able to successfully make requests without it. * **"Secret"** - This is a secret key. Every account will have a unique value that you will need to use for this field. From the Riku dashboard, look at the sidebar where we have the option to "Manage Account". Click this and you will see a popup with your secret key. This is also a required field for successful requests. Do not share your secret key with anyone! * **"Prompt ID"** - This is the prompt you are looking to call. We covered this a bit earlier so you can read about how to get your Prompt ID [here](/exporting-prompts/single-endpoint-export/getting-prompt-ids). This is also a required field so please ensure it is sent with all requests to avoid errors. * **"Input 1"** - If the prompt you are wanting to use has an input field, this would be the first of those input fields. For example, a Twitter Hashtag Generator prompt would have a single input for the tweet. This is the field where you would put that tweet. * **"Input 2"** - This is the second input field. If you are using a prompt with 2 or more input fields then you would be required to enter something here. An example would be a product description generator having two fields of Product Title and Product Details, the second of these inputs is the Product Details and what you would put to correspond here. * **"Input 3"** - This would be used if there are three or more inputs and works in the same way as specified above. * **"Input 4"** - This would be used if there are four or more inputs and works in the same way as specified above. * **"Input 5"** - This would be used if there are five inputs and works in the same way as specified above. * **"n"** - This field is telling us how many outputs you are expecting to achieve. A value of 1 will return a single output whereas a value of 4 would return 4 outputs. Note that if the prompt is using a large number of tokens then the ability to use the n might be limited. If you see errors after setting this value too high, try setting it lower. # Full Example Export On this page, we'll show a full export showing the steps we like to use. You will also learn how you can get the different code for NodeJS, Python, and many more methods. We will go back to the board game example we showcased when we were getting the Prompt ID and can put that Prompt ID into our request. ``` curl --location --request POST 'https://prompts.riku.ai/webhook/run' \ --header 'Content-Type: application/json' \ --data-raw '{ "Name": "YOUR NAME", "Secret": "YOUR SECRET", "Prompt ID": "1649450161206x134604324816551940", "n": 4 }' ``` With this request, as there are no input fields - they have been removed. You will need to add in your details in the Name and Secret fields and you can then copy this and open up Postman and hit the Import button to Import this into Postman. You will need to select Raw and follow the simple steps. ![](/files/rA6GMFL422CUCEefUuqP) Awesome! Now we have it in Postman successfully. If we wanted to test that it works. We can just hit the blue "Send" button in Postman to confirm we get the 4 outputs as specified in our "n" field. You can modify this number if you like to see how it affects the outputs. ![](/files/se0aafSfi0k3lTcV7kjs) Super! It works as intended and returns us an array of the outputs which we can then use in our own workflows. If I wanted to get the code for Python or other places, I can do that super easily from Postman. Here is how to do that. ![](/files/TPEnaBLFjjrWKjgGbQI8) So there we have it! From a single endpoint, you can use all of the best AI technology and we do all of that routing for you to ensure it goes to the right place. You just need to change the Prompt ID and ensure you have the right input fields set each time. It could not be easier! We're excited to see what you build using Riku! # OpenAI The company with the deep pockets and great models. OpenAI can be credited with making AI and ML trendy and bringing a lot more people to the space. People got very excited about GPT-2 when it was released publicly and with the private beta of GPT-3 from 2020 onwards, developers have been building on top of the technology in a rapid speed. GPT-3 from OpenAI offers a few different models. The most powerful of these is the davinci model which has 175 billion parameters. The other models include curie, babbage and ada. OpenAI have also released a code specific model which has a lot of knowledge of github. The OpenAI infrastructure is reliable and easy to build on. They offer an online playground where you can create and interact with the models without using the APIs. OpenAI take a very strict rule on what they will allow you to build and every single user has to go through a pre-launch review process prior to introducing features to their audience. There are also some completely off limit use cases such as automated social media, anything to do with Instagram or Twitter and Adult content. If you are looking to build a business application then OpenAI is a good choice. The davinci model can be expensive if you are looking to generate large amounts of text whereas the curie model is more reasonable and provides good outputs with quality prompt engineering. Like everything, it is worth experimenting for your specific usecase. Fine-tuning is an area of focus where OpenAI are starting to lead the way. Being able to create a fine-tune with your data will help you get a more relevant outcome and being able to self-serve your fine-tune needs is super important to reduce friction and allow those who have the ability to master their own datasets to get the most out of this technology. With extremely limited exceptions, OpenAI requires all production live usecases to have their content filter active. The filter returns a 0, 1 or 2. If it returns a 2, you are not supposed to provide the end-user with that output. The filter can be frustratingly broad at times and is a definite downside to using OpenAI’s technology. OpenAI recently took their models out of private beta and into a public beta and that means that anyone can go to their website and signup. This helps to bring the ecosystem into the open as prior to this you could find yourself waiting months on a waiting list. I waited 5 months personally to get access back in 2020! # Cohere This Toronto company has come out of stealth with a wealth of funding and a founding team with some serious clout. The team have a close relationship with Google and some of the founders worked on some of the most prominent AI/ML projects within Google prior to starting this company. Cohere offer a bunch of tools to help with your AI needs and don’t just focus on generating text. You can use their models for sentiment analysis and figuring out patterns. When talking with the team, you get a sense of their deep understanding and passion for this space so it is definitely worth keeping a track of what they are up to. Cohere offer 3 models for text generation, a small, medium and large. They don’t disclose the size of the models in terms of parameters but they seem well trained and the API is easy to use. The pricing is also very generous compared to others on the market. One of the areas of Cohere which I think is above anyone else currently is the way that you are able to fine-tune the models and create these yourself in a fully self-serve scenario. This process is simple, fast and get you the API endpoint immediately to implement into production. As far as prompts go with Cohere, I have found that you need to phrase things slightly differently. Whereas a prompt used in an OpenAI model can be pasted into an AI21 model and work just as well, there is a definite art to using cohere which can be a little frustrating and might put some people off. Cohere take an approach similar to OpenAI with ethics and morals of the technology wanting all users to go through a launch review prior to putting the technology into production. There are some banned use cases such as not creating full blog posts etc which you should be aware of. Also through testing, there is a filter on the models which is built into the API endpoints. This is a better system than the one from OpenAI but a filter is still frustrating for certain things such as my testing with a female ecommerce fashion brand which kept hitting it for skirts and underwear descriptions. The price of Cohere really is a strong reason to include it in your AI stack of tools. The team are approachable and willing to help you out and I believe that they will constantly improve the models. The fine-tuning definitely is where the product comes alive. If you have a dataset of over 1MB and want a fine-tune, you will not find an easier way to get the fine-tune done. # AI21 This Israeli company announced their models in 2021. They also have a successful product in Wordtune which is a rephraser and one of the best on the market competing primarily with quillbot. AI21 and their Jurassic models are a similar size to OpenAI’s davinci and curie. AI21 are the new kids on the block and have significant backing. They launched with a focus on fine-tuning and a comfy Discord for users. The interface on the website in the studio is very similar to the OpenAI playground so you can create and play with the models here. The API is easy to use and the pricing is relatively simple. One of the major advantages of AI21 over any of the other models on the market is that AI21 only charges for output tokens. This monetization policy is positive for a few use cases; prompt engineering becomes easier as you don’t have to balance out examples and cost. You can fill up the tokens with examples as you only pay for the output. This monetization policy is also incredibly good for tools which have a ‘Write for Me’ function as you are not paying for the text in the prompt every single time the end-user hits generate. Fine-tuning with AI21 can be a slower process due to the fact that you have to provide their team with your files and information and they do it manually on their side. They are working on a way to make this self-serve which will be an incredible breakthrough but for the moment, this is a bit of a downside. There is no filter on the AI21 models which is great for creating content that might get censored with other technology. The process for going live is also incredibly simple. Competition is always great at driving down prices and forcing companies to innovate and AI21’s introduction to the market has definitely had an impact on some of the more dominant players. # EleutherAI EleutherAI is really an outlier in the model space as everything that they create is open source. You can host their models yourself on your own device and have the power of this AI technology without any filters, oversight and use it as you wish. They have been producing better and better quality models with GPT-Neo and GPT-J getting the most fans. Whilst these models are open source, they are billions of parameters of data so you need to have a pretty hefty device to be able to run them efficiently and multiple requests at once can really cause problems. Unless you are willing to spend a ton of money in supporting this infrastructure, the open source dream is sadly out of reach for many. What we like about EleutherAI is that they have a thriving Discord community and really clever people working on solving some of the biggest AI problems with no ulterior motive than to forward the movement and make the tech more accessible and less dominated by a few players. That is to be commended! The models themselves are trained on quality data and perform incredibly well across most tests so for any serios AI afficionado, they should be part of your stack! # Aleph Alpha *Coming soon* # Muse API *Coming soon* # Introduction to Fine-tuning Fine-tuning often confuses people. It doesn't need to be complicated and we're here to put it in simple terms. Fine-tuning is the process of taking a larger dataset and formulating it into a prompt so that you can get better outputs from the AI models. Think of fine-tuning like building out a prompt but on a much larger scale. With fine-tuning you are no longer limited to a maximum token amount so you can provide a much larger sample of examples than you can with just a prompt. As an example, imagine you are building out a blog introduction prompt and you are giving a few examples and end up reaching the token limit. You may be able to include perhaps 5 or 10 of these examples maximum. It will give the AI a good opportunity to learn the pattern and the stronger the underlying AI model is, the better the output will be. Often times, this is enough to get an output you are happy with but sometimes for more complicated scenarios, you may want to consider fine-tuning. If we use the same example and go into how fine-tuning would work, instead of providing just 5 or 10 examples, we could provide 100, 500, or 10,000 examples. Considerably more data and more training for the AI to get a deeper understanding of the content and the type of output that we are expecting as an output. By providing these larger datasets, you are going to get a better model that performs in a whole new level compared to just a vanilla AI model. # Importing from OpenAI OpenAI allow you to create fine-tunes for all of their standard models. They don't have a user-friendly way to do this currently and it is only available to be done via code. We are working on building a no-code friendly UX for this and it should be out soon but for now, if you have created a fine-tune with OpenAI, you can import those models into Riku for your use. To get started with importing your OpenAI fine-tuned models into Riku, you will need to go to the OpenAI dashboard which you can get to [here](https://beta.openai.com/playground). Once you are in the Playground of OpenAI, you can click the Engine dropdown to see all of the OpenAI models. If you scroll down to the bottom you will see "Fine-tunes" where you can see all of the fine-tunes that are available in your account. ![](/files/ohx2DAFjAHdX4btpuKcT) Select the model that you would like to import into Riku and then you will be able to click the view code on the top tool bar. From here, I like to make it so it is showing curl, it might show python by default so you can change this to curl by clicking on that and selecting curl. ![](/files/yU2isXxeUdiNdlQBfKMp) You can see in this example that there is a fine-tuned model and we are interested in copying everything within the quotation marks shown above. Copy this to your clipboard and load up Riku. Within Riku, click on your API Settings tab or directly click [here](https://riku.ai/dashboard/api-settings). See the Fine-tune Models option and select OpenAI from the dropdown and hit the button to load the popup. ![](/files/mke0hBI9AYYR1IvkCVQ0) Write a name for the fine-tune that you recognize, this is just internal for you and for you to remember what the model does and nothing more. Then take the fine-tune ID that we got previously from OpenAI and enter it into the second input. Hit save and you will see a success message. If you then go to the Playground in Riku and change the technology to Fine-tuned Models you will see that the new fine-tune will appear and you can select it. ![](/files/cxFH9VN7FcU5mALwQEyT) There we have it! It is that simple to add an OpenAI fine-tune to your RIku account. You can then use that model in the playground, create saved prompts with it and even create public share links for others to use. I hope this guide has been useful! # Importing from Cohere Cohere allow you to create fine-tunes for all of their standard models. You can currently do this on the Cohere website by selecting a file and uploading it, code isn't required. We are working on building a no-code friendly UX for this directly within Riku but for now, if you have created a fine-tune with Cohere, you can import those models into Riku for your use. To get started with importing your Cohere fine-tuned models into Riku, you will need to go to the Cohere dashboard which you can get to [here](https://os.cohere.ai/). Once you are in the dashboard of Cohere, you can see the ready models. Look for the fine-tuned model which you want to import into Riku. ![We will be demonstrating how to import this sectionwriterdataset.](/files/RPZ4PEFWs0XvusnjZObi) The top model here is a fine-tune so we will be wanting to get the ID so we can put it into RIku. If we click on API Link, we will see a notification that something has been copied to the clipboard. For this example, my clipboard looks like this - . I want to remove everything up to the .ai/ so I am only interested in this string - 28a32b25-c0f6-4426-b38e-b6f4463723e7-ft. Keep this copied as we will now go to Riku to enter it. Within Riku, click on your API Settings tab or directly click [here](https://riku.ai/dashboard/api-settings). See the Fine-tune Models option and select Cohere from the dropdown and hit the button to load the popup. ![](/files/8btxZH0hJM617lUDDNKD) Write a name for the fine-tune that you recognize, this is just internal for you and for you to remember what the model does and nothing more. Then take the fine-tune ID that we got previously from Cohere and enter it into the second input. Hit save and you will see a success message. If you then go to the Playground in Riku and change the technology to Fine-tuned Models you will see that the new fine-tune will appear and you can select it. There we have it! It is that simple to add an Cohere fine-tune to your RIku account. You can then use that model in the playground, create saved prompts with it and even create public share links for others to use. I hope this guide has been useful! # Importing from AI21 AI21 allow you to create fine-tunes for all of their standard models. You can currently do this on the Cohere website by selecting a file and uploading it, code isn't required. We are working on building a no-code friendly UX for this directly within Riku but for now, if you have created a fine-tune with AI21, you can import those models into Riku for your use. To get started with importing your AI21 fine-tuned models into Riku, you will need to go to the AI21 dashboard which you can get to [here](https://studio.ai21.com/playground). Once you are in the dashboard of AI21, you can see the models in the menu on the left. Look for the fine-tuned model which you want to import into Riku and select it. ![For this example, we will take the blog-writer-v2](/files/Cph5H3wF6zVAWESrwU4A) Once we have selected the model which we want to import, we can click the API button in the top of the AI21 dashboard. We can then change this to curl and we can look at the URL. We want to take note of two things here. Firstly, we want to copy the fine-tune id. This is the string before /complete. We also want to note down which of the AI21 models this has been fine-tuned with. For this example, the answer is j1-large. ![Copy the string before /complete and take note of which AI21 model was used.](/files/5wqLPhSYroRZuezR4zdV) Within Riku, click on your API Settings tab or directly click [here](https://riku.ai/dashboard/api-settings). See the Fine-tune Models option and select AI21 from the dropdown and hit the button to load the popup. ![](/files/26lFQMEJLpoALJKSpkuc) Write a name for the fine-tune that you recognize, this is just internal for you and for you to remember what the model does and nothing more. Then take the fine-tune ID that we got previously from AI21 and enter it into the second input. Finally, we need to put in the AI21 model which we noted in the previous step. When you are sure you have entered everything correctly, hit save. If you then go to the Playground in Riku and change the technology to Fine-tuned Models you will see that the new fine-tune will appear and you can select it. There we have it! It is that simple to add an AI21 fine-tune to your RIku account. You can then use that model in the playground, create saved prompts with it and even create public share links for others to use. I hope this guide has been useful! # Creating Prompts Prompts are central to everything that we do with Riku. We want you to be comfortable with experimenting and making the prompts that you need for your business. There is a bit of a learning curve when it comes to making a prompt, we hope to make that as painless as possible though if you just read everything on this page. Once you are logged into your Riku.ai account, you can access the playground [here](https://riku.ai/dashboard/playground). The playground is where you can experiment with the raw AI models and get something great out the other end. There are various different settings you can alter. {% hint style="info" %} Hovering over any of the settings in the right sidebar on the playground page will give you a brief overview of what that setting does. It is easy to forget so use this information when building out a prompt. {% endhint %} Generally the steps we follow for making a good prompt are the following; 1. Give the AI a clear instruction at the top. This could be anything that you want it to output. An example could be; "Write me an introduction for a blog post on the following topic which is exciting and engaging." The more details you write in this instruction, the better. 2. Think about what your inputs should be. What would the end-user or you like to put in as information for the AI to work with. This could be a single input or multiple inputs. If we follow on for a blog introduction, we probably want 2 fields; Blog Name and Blog Description. You can either label these as such to help you remember or shorten them down so you could put BN: and then BD: on the next line, or you could write them out in full. We like to keep the inputs with a colon next to them to help the AI learn the pattern. 3. What we'd do now is actually make an example or two of these inputs. We'd fill in all the input fields we have created and then on the next line, we'd put a differentiator. By default and what we feel is good practice is to use '###' as the likelihood of you using this sequence in your writing is pretty low. 4. On the next line after the ###, you can give your output example. Write in the best way you can, a blog introduction based on the information you have provided for the inputs. When you are done, on the next line put ###. Now the AI will know that the pattern we are following is inputs, separator, output, separator and then a new sequence will occur. 5. Congratulations, you have now created a prompt with a single example. You can make this prompt better by providing further examples, how many is entirely up to you. You can also test it by providing the inputs again, and following with the ### and then hitting generate to see what comes out. You may want to then edit the output, toggle some of the settings, change AI engine or a mixture of them all. 6. When you have a prompt you are happy with, you can hit the save button. Make sure that the prompt finishes with your separator so it should be your final example inputs, your separator (###), your final example output, final separator (###). When saving, follow the fields and enter the information. You can choose whether to make the model public to the community or save it as a private model for your account.
# Saving & Sharing Prompts Once you have [created a prompt](/rikus-playground/creating-prompts) you can save it to either have in your private collection or to share with the community. There is a purple button with 'Save Prompt' on it. Click this button and you will be asked to give the model a name and also to provide the number of input fields. If you are asking the end-user to give one detail before each output, you'd put 1, if there was more then you'd put that number. In Riku.ai, the maximum number of inputs is set as 5. You will then be able to define the inputs. There are three fields to fill in for each input you specified. The first field is the input as it appears in the prompt, for example "Blog Name:", please make sure to provide the input as it is written and also the colon. The second field is how the input will appear on the front-end when you call it within Riku.ai, for this example; "What is your blog called?" would work. The final field for an input is an example of it being used so we could either use one of the examples we added to the prompt or something new. If we were doing a blog on AI Copywriting then we might want to add as the example for the third field "Why use an AI Copywriter?". {% hint style="info" %} Next up is adding a description for the prompt. This will be helpful for members of the community to determine whether they want to use your prompt in the future. It will also give you memory of what the prompt is all about if you are not making the prompt public. Provide 30-50 words ideally for the prompt description. {% endhint %} The final step is deciding whether to make the prompt available in the community showcase or keeping it private within your account. You are in full control here. Saving a prompt for the community will let other members see the examples, design and settings you have chosen. Keeping it private means that only you can access it. Expect a delay of up to 6 hours on community prompts being approved. We tend to approve them all but have to be vigilant for bad faith actors. # Community Showcase The community showcase is a place to show off the latest prompts and find something that you need. These are all the best prompts from the community and we hope that the best quality prompts rise to the top. You can see the prompt titles, description, which AI model they use and who created them. Clicking on any of these options will load up a place where you can run that prompt with your own inputs. You are also able to edit any of these models to swap out inputs for something which might be better suited for your needs and business. ![](/files/Bpt2FJ4bGQDpw6n3j08s) There is a search bar at the top where you can write out what you are looking for to filter just by those prompts. If you are wanting to write a blog for example, you could search with the term blog and it will show all the blog specific prompts from the community. There is a second search on the right side where you can filter by technology. # Private Prompts Your private prompts can be accessed by clicking the option in the left sidebar. Once you click this, you will be able to see all the prompts you have saved to your account. You will see prompts you have saved and not made public to others and you will also see your community prompts here. Everything created by you and saved is shown here. ![](/files/rob67Ym2Pu2E9piAdTSH) Like with the community showcase, click on any of the prompts and you will be taken to a screen where you can provide inputs and execute them within your account. This is your little private area to enjoy the vault of all your creations! # Execute Prompts All of the prompts within Riku.ai can be executed directly within Riku! You can go to either the Community Showcase or Private Prompts and choose what you'd like to run. Click it and it will load up the execution page. You will see the input titles and examples of text to input. Fill in these fields and the final option is to choose how many you would like to create from 1 to 4. ![](/files/vzEBkpSFaUUIXry7GjeH) Once you have the outputs, you can choose to use them for your projects, you can choose to edit the prompt in the playground to fit it more to your needs or you can go back to the input fields and tweak them slightly. We want you to have the best experience with AI. # Code Export Our code export process has changed and now is detailed in the Single Endpoint Export. {% content-ref url="/pages/zLq8aXo4pbTr9jL0ZcxC" %} [Single Endpoint Export](/exporting-prompts/single-endpoint-export) {% endcontent-ref %}