Skip to main content

Merge Agent System Prompt with Project System Prompt when Both are Set

Feature Requests

Currently, only one system prompt can be active at a time — either an Agent System Prompt or a Project System Prompt.

It would be great to support merging these. For example, I have a project-level system prompt that instructs the model to access certain information via MCP as base context. I'd like to use this project prompt in combination with different agent prompts simultaneously (e.g., Business Consultant, Technical Consultant, etc.).

Posted by Maximilian Christl ·
Status: Completed7 comments

Log in to comment and vote

Comments7

  • Daniel Nguyen

    Team•

    Apr 28

    I’ve added this in v2.11. It’s trickier than just “merging agent prompt with project prompt”. The reason is that when we change the chat from agent profile to custom chat profile, it’s can be awkward not to merge this custom chat’s system prompt with the project.

    So I designed it so that the Project’s prompt will be the “base prompt”. Any customization in the chat (whether comes from agent, or the chat itself) will be merged with this base prompt.

  • Maximilian Christl

    •

    Mar 15

    Hi @Daniel Nguyen , I just wanted to renew this request. Even a simple string concatenation that could be activated via a checkbox would be great. Currently, one has to decide whether to use either Agents or Projects. No real combination is possible, as the system prompt of the Project would get entirely lost. It would be enough to simply merge the Project's prompt after the Agent's prompt.

    • Daniel Nguyen

      Team•

      Apr 25

      Working on this. Sorry for the delay.

  • Dustin Weßels

    •

    Feb 4

    If it is any help, I was just struggling with this too, would love this feature.

  • Maximilian Christl

    •

    Jan 5

    Hi Daniel,

    Thanks for your reply—I love the idea. I've been thinking about developing a system myself that clearly separates the following structures, but BoltAi is already ticking many boxes:

    • Model Configurations

    • System Prompts (Agent, Base Prompt, etc.)

    • Contexts (Files, additional .md or .txt instructions, etc.)

    • Knowledge Bases (more like a database or embedding store that can be accessed on demand, similar to how Cherry Studio handles this)

    • Memories

    • Tools and Plugins

    All of these should be individually created and managed, but also composable—meaning you can combine them for any chat or within Projects, which would be combinations of the structures above.

    I'm curious whether your concept is heading in a similar direction. Just an additional thought.

  • Daniel Nguyen

    Team•

    Jan 5

    Yeah. I have a plan for this feature. Other users want the ability to “inherit” system prompt too.

    But I think it’s very hard to maintain them this way.

    My solution is to allow compose them, instead of inheritance. If you’re a developer, you probably have heard of the “Composition over inheritance” concept.

    Basically, UX wise I’ll add the ability to use @ to mention/tag another entity: it can be a prompt you configured, another agent, project prompt etc.

  • Maximilian Christl

    •

    Jan 4

    One additional note: as a workaround, you can currently add these instructions as a file, though this requires using an external editor. Perhaps another option would be to include a built-in text editor for adding context files, allowing you to provide additional synced context to projects or agents directly within the application.