Enza's Trash Trove
Table of Contents
- Mistral Guide (on Janitor AI)
- First Steps
- Getting your API key
- Choosing your model
- Putting everything in your configuration
- Frequently Asked Questions
- What is the limit for the free tier?
- Mistral is outputting weird/bad replies!
- I’m encountering × error, what do I do?
- Mistral is saying gibberish!/Mistral is repeating itself!/What temperature should I use?
- Do I have to pay?
- Is it better than Deepseek?
- The bot is responding with a bunch of [object Object]s!
- Why should I use Mistral over Gemini/Deepseek?
- Mistral Prompt (co-authored by Sprout!)
- Other information and tools
Mistral Guide (on Janitor AI)
Use Mistral as a last ditch effort if you cannot use any other alternative.
First Steps
Go to your Mistral API Plans, and upgrade to the "Experiment" plan.
This requires a phone number, but you can use a service like Google Voice or TextNow, which Mistral allows.
I personally have not used these services and cannot endorse them.
Getting your API key
Go to Mistral’s Console and create an API through there. You can name it if you wish.
Wait for your API key to be validated, this usually takes 1-5 minutes.
Choosing your model
Next, you’re gonna choose a model. Here (and here!) is a list of available models, copy one of the model names (e.g. magistral-medium-latest).
If you’re seeing the prices don’t worry! You won’t be charged as long as you’re using the free experimental tier!
Your choice of model is usually up to preference, and as long as you're using one of the latest models (Mistral Small Creative, Magistral Medium, etc) you should be fine. I personally switch between the two depending on my mood and need.
Putting everything in your configuration
Lastly, we’ll go to Janitor.AI to put in all our details!
- Put your chosen model's name in "Model Name"
- Put this in "Proxy URL", DO NOT OPEN IT:
https://api.mistral.ai/v1/chat/completions - Put your API key from the Console earlier
Refresh the page and check if your API Key/Model are working, then you should be good to go!
Frequently Asked Questions
What is the limit for the free tier?
Mistral offers 500000 Tokens per Minute and 1000000000 Tokens per Month for free! That’s a shit ton! Learn more here!
Mistral is outputting weird/bad replies!
Change your prompt! There are a bunch of prompts you can choose, but I've also made my own prompt (with help from Sprout!) here, it's more recent and is specifically catered to Mistral!
I’m encountering × error, what do I do?
So far there are these are the known errors;
401
This error occurs when the API key is invalid or missing. Double-check your API key! This could also mean that your API key hasn't been validated yet, wait a few minutes before trying again!
404
This means something is incorrect in your configuration! Check your model name, URL, and API key again to see if they're correct.
429
This error indicates that you have exceeded the API’s rate limit. This can be either the Requests/Second, Tokens/Minute, and rarely, the Tokens/Month limit.
Mistral is saying gibberish!/Mistral is repeating itself!/What temperature should I use?
Magistral is most behaved at or below 0.7!
Other models (those that are non-reasoning) run best at high temps, think 1.3 or higher!
If it's repeating itself, increase by 0.1! If it's saying gibberish, decrease by 0.1! Try what suits you best!
Do I have to pay?
No! That's why it says in the title free!
Is it better than Deepseek?
Nope. But it's better than most (completely) free alternatives out there, and the newer models are pretty good once you get them set up.
The bot is responding with a bunch of [object Object]s!
This is an unconfirmed error, but is speculated to be caused by a rate limit/quota error. This is an error that indicates a rate limit imposed on you, but instead of an error code it posts it into the chat. The only solution is to reroll or wait until you get your message. If a reroll does not fix it, try switching to a different model (e.g. to Magistral Small 1.1 or a non-reasoning model).
Why should I use Mistral over Gemini/Deepseek?
You shouldn't as long as you still have your free messages. It is recommended that you choose a better model (like from the R1 Gemini, even Haiku (lol) families of models). Mistral is mostly a backup for when you can't access SotA (state of the art) models.
Mistral Prompt (co-authored by Sprout!)
Here is the prompt for non-thinking models, used best with Mistral NeMo and Medium:
I am a co-author in {{user}}'s story! My job is to collaborate with {{user}} to create a narrative that is engaging, unpredictable, and full of life!
I am in control of {{char}}'s thoughts and actions, and also any characters that might come along the way! I always write with {{char}} in mind, thinking about their feelings, wants, and needs with every response. {{user}} is an independent character that I cannot control; I only have authority to control other characters in the story.
I will be bold and spontaneous in my writing! I won’t just wait for {{user}} to lead; I will take the initiative to introduce unexpected plot twists, new environmental shifts, or sudden events that keep the story fresh. I will avoid predictable clichés, choosing instead to surprise the reader with complex character reactions and hidden motives.
I will write long, elaborate prose filled with sensory details, using metaphors and evocative language to heighten the immersion. I’ll keep the reader’s heart racing by using suspense and revealing secrets slowly! No matter the scene, I will match the tone perfectly—being depressing when it’s sad or elated when it’s happy—while ensuring every action feels natural and raw.
Optional formatting—for those who prefer different formatting styles feel feel to change this:
I will follow these specific formatting rules: Dialogue: Use quotation marks (" "). Meaningful Pauses: Use em dashes (—). Emphasis: Use italics (* *). Heavy Emphasis: Use bolding (** **).
Add this if you're using a reasoning model (Magistral) to format it:
I will remember to start my reasoning with <think>, and end it with </think> before starting my actual response! Since there is no need, I will also not provide summaries!