Custom instructions
Custom instructions let each user define personal context that is automatically injected into the system prompt of every Assistant they chat with — similar to ChatGPT's Custom Instructions. They are useful for telling the model who you are and how you prefer it to respond, without having to repeat it in every conversation.
Where to find them
Custom instructions live in your personal profile, under User profile → AI.
The form has three parts:
| Field | Description |
|---|---|
| Active | A toggle to temporarily enable or disable your custom instructions without losing their content. |
| About you | What you would like the model to know about you to provide better responses. |
| Model instructions | How you would like the model to respond (tone, format, level of detail…). |
Each text field accepts up to 1500 characters and shows a live character counter. Changes are saved with the Save button.
The AI section of your profile also has Limits and Consumption tabs, where you can track your AI spending per organization and project — see AI Cost Limits.
How they are applied
- When the toggle is active and at least one field has content, the text is injected into the Assistant's system prompt on every execution.
- They are treated as personal preferences: the Assistant's own configured instructions and tool rules take priority, so custom instructions cannot override its task.
- They apply across all providers (OpenAI, Anthropic, Google, Azure).
- They apply to all threads you create, and only to yours — other users' threads are not affected.
- When the toggle is off, nothing is injected.
Because custom instructions are injected into the system prompt, they can change how Assistants behave — their responses, tone, and decisions. If an Assistant starts behaving unexpectedly, check whether your custom instructions are active.
See also
- Threads — where your custom instructions take effect.
- Assistants — the configured Instructions that always take priority over your personal ones.
- AI Cost Limits — track your AI spend per organization and project.