Update on instructions feature in chat settings

Continuing the discussion from Best practices/Suggestions for "Knowledge" instructions:

Ahem!, been a while since I last posted and thought I’ll update the handbook for one of my older posts as the interface for adding instructions have also changed.

tl;dr: Check the tl;dr: of the first post :eyes:

So… without further ado, here’s where you can find the new placement to add instructions:

The v0 team has also added an option that allows the agent to automatically create images, in places when needed or instructed, by simply toggling the “Generate Images” button at the top. They have also added support for MCP integration, you can find this in another blog or I may create one in the future.

Once you click on this “Instructions” menu item, it’ll open a sub-menu listing all your instructions with an option to filter using search input and another to add new instruction:

By default, you get the “Be Concise” and “Plan Mode” instructions which are both very great throughout project development. To add a new instruction, simply click on the “New instruction” item and it’ll open a dialog as follows:

In the title, you can add something that helps you identify your new instruction and in the rule, you should mention what you want the agent to do in which scenario. Try to keep it short and clear, maybe use some other AI model and have it generate the rule for you. Let the model know what you want the rule to be for and ask it to keep it short but properly define the expected output. For instance, one of the rules I’ve set when developing mobile focused apps:

Now humans are known to make mistakes, so the developers at Vercel have added an option to edit your rules that can be accessed from the Instructions sub-menu, by clicking on the filter icon that appears next to the title of your created instruction:

They also support the deletion of instructions through the dialog that pops up when creating or editing your instructions.

That’s all for this discussion, let me know your thoughts and some instructions that have helped you the most with guiding the v0 agent to provide better outputs. Peace! :victory_hand:

Happy Holidays!

5 Likes

Thanks for the detailed update post :handshake:

I’m sure lots of folks will find it useful.

2 Likes

Was hoping someone else does it, but didn’t find any post that did. Hope it helps! :peace_symbol:

3 Likes

Thanks for this - I recently couldn’t find where my ‘project instructions’ had gone, and needed to add more to it.
It looks like you’ve separated some of your instructions from the earlier post into different selectable instructions.
Configurable, but I thought recreating that myself might be tedious.

Wanted to share that I took a slightly different approach - one that enables versioning and migration (potentially) to another AI, and because for the most part I want all my instructions followed all the time.

I had v0 source my original project instructions, review history for new learnings instructions, and then create an '/AI_CONTEXT.md’ file.
Then I created a single instruction ‘Always read /AI_CONTEXT.md at the start of each conversation before making any changes.’

A little simpler to set up and manage, if you don’t mind the ‘all or nothing’ baseline this creates.
I can pull things out separately later if I feel the need.

Thanks again - this, and your prior, post helped a lot.

1 Like

Glad to be of help!
Out of curiosity, can you check if it costs you more while prompting with this instruction on? Because you’re adding a file on every prompt, it might get expensive if you’re adding to the file over time. Not 100% sure, will have to ask the pros!

Good question, no idea.
I’d have thought there’d be little difference in the tokens used to ‘read’ or follow project rules or selected instructions every prompt, vs reading the file - but it’s a fair callout.
I’ll try to pay attention and see if I notice a difference.
Of course, there’s benefit/upside if it means 2-3 prompts to get something done rather than 10.
I’ll see.

EDIT: v0 itself says it will consume more

1 Like