Add LLM-specific instructions
In addition to global instructions
Currently, we can add agent-specific instructions, so another way to implement this is:
https://feedback.typingmind.com/p/select-an-agent-in-multi-llm-chats
In addition to global instructions
Currently, we can add agent-specific instructions, so another way to implement this is:
https://feedback.typingmind.com/p/select-an-agent-in-multi-llm-chats
Log in to comment and vote
Comments5
Tan Crocodile
Sep 28, 2025
•Merged request
•2 votes
System intruction
Model configurations (add/edit) should have an optional field for adding to or replacing the TM global System instruction thing.
Ivory Alpaca
Sep 16, 2025
related:
https://feedback.typingmind.com/p/add-llm-specific-instructions
Blue Raccoon
Sep 18, 2025
I would advocate for a more comprehensive yet clear schema for default prompting:
One possible example:
—> Instance (for team version only)
—> User
—> Model
—> Agent
—> Project
With supporting tools to view the resulting prepend for any given chat, allow lower levels to over-ride prior level default prompting, etc.
I recognize that the system supports most of this in some ways today, but increased clarity around order of operations, and better UI and tools for management and troubleshooting would be very helpful.
Ivory Alpaca
Sep 7, 2025
A certain LLM might need specific instructions. For example, I consistently find Grok models too verbose. So, I will add instructions like "Distill your replies, adjust the length according to the complexity of the query, definitely < 1000 words, < 500 if possible (for math problem and thinking mode, it's fine to go > 500; for Fast (non-thinking) mode, ideally < 300, and 500 is the hard limit); when the distillation is hard, ask constructive questions and gather feedback for a full answer. "
Ivory Alpaca
Sep 7, 2025
Related features not enough:
Feature
Issue
Agents
Cannot be used in multi-LLM chats; limited to single-model contexts.
Projects
Apply global instructions to all models in a project; no per-model differentiation.
Global Prompts
Requires verbose conditionals (
if-elselogic), wasting tokens and increasing brittleness.