Configurable Default System Prompt for Port AI Assistant
under review
D
Daniel Kosalla
The default Port AI Assistant does not support a configurable system prompt. This causes inconsistent AI behavior across organizations — the AI may ignore org-specific docs, use incorrect tool calls for custom blueprints, or produce poorly formatted output for specific interfaces (e.g., Slack).
Current workarounds (AI Chat Widget, custom Agents, Skills) are either scoped to specific dashboards or require manual user action, making them insufficient for org-wide consistency.
Proposed Solution
Add an organization-level default system prompt setting (e.g., under Organization Settings → AI) that allows platform engineers to:
Override the default system prompt for the Port AI Assistant globally.
Control which Skills are always loaded or excluded per interface.
Define interface-specific prompts (e.g., web vs. Slack).
Expected Outcome
Consistent, context-aware AI behavior across all interfaces — without requiring users to manually select agents or craft explicit queries.
Sagi Keren
Merged in a post:
Customize default Port AI side chat agent
J
Justin Wied
Our users like the ability to click the circle with stars in the lower right and begin a chat with the Port AI. Nearly 100% of the time they are asking about data within our tenant/org.
Request - It would be helpful to be able to customize the default AI agent that users talk with, to give it more context and tell it how to respond to certain questions. This would allow for the AI agent to give much better, contextual responses to the everyday Port user.
Sagi Keren
updated the status to
under review
Matan Grady
Merged in a post:
Configurable Default System Prompt for Port AI Assistant
D
Daniel Kosalla
Problem
Without server-level instructions, the Port MCP Server lacks organizational context by default, leading to undeterministic responses, incorrect tool calls, and agents unfamiliar with internal terminology
Existing workarounds, Skills, custom prompts, and separate agents, all require extra distribution and setup friction that server-level instructions would eliminate entirely.
Proposed Solution
Allow platform engineers to define custom instructions at the MCP Server level that are:
Injected as the first point of context on every interaction, bypassing the need for subsequent reasoning loops
Configurable once via Port UI/API, applying universally to all developers without per-user setup
Able to encode internal terminology, blueprint priorities, and organizational guidance
Expected Outcome
Consistent, terminology-aware AI responses across the organization
Zero distribution friction compared to Skills or custom prompts
Reduced need for developers to be explicit in every query