Skip to main content

Agent Builder

Status

Owner: TBD · Status: draft · Canonical source: flezi.app.frontend.main/src/components/agent-builder/

Who is this for?​

AI solution builders and advanced users (AI_BUILDER, WORKSPACE_ADMIN).

What can you do?​

  • Define system instructions and behavioral guardrails.
  • Select foundation models (GPT-4o, Claude 3.5 Sonnet, GPT-4o-mini).
  • Bind tools, skills, knowledge bases, and MCP connectors.
  • Configure subagent delegation hierarchies.
  • Attach AGT security policies and conduct live dry-runs.
System Instructions and MCP bindings

System Instructions add guidance but do not erase Flezi identity or safety precedence. Users or administrators select MCP connectors from the catalog; Agent Builder does not invent MCP bindings. The session capability surface contains only the bound tools, skills, and MCP connectors.

Steps​

  1. In the Agent details view, select the Builder tab (or Build with AI).
  2. Set System Instructions: Describe roles, objectives, and constraints.
  3. Under Model Selection: Choose the model and adjust temperature.
  4. Under Capabilities & Tools: Enable sandbox tools, web browsing, or RAG search.
  5. Under Security Policy: Select an AGT Policy to govern tool calls.
  6. Use the Test Agent panel on the right to interact and refine before publishing.