Environment Configuration

Environment configurations define tools, resources, and prompts that are available during agent execution. Unlike Knowledge sources (which are loaded at startup), environment resources and prompts are loaded on-demand when tools request them.

Example Configuration

agents:
  code_analyzer:
    environment:
      type: "inline"  # or "file" to load from separate file

      # Tool Definitions
      tools:
        analyze_complexity:
          import_path: "myapp.tools:analyze_complexity"
          description: "Analyze code complexity"

        git_blame:
          import_path: "myapp.git:blame"
          description: "Get git blame information"
          requires_confirmation: true

        format_code:
          import_path: "black.format_str"
          description: "Format Python code using black"
          settings:
            line_length: 88

      # On-demand Resources
      resources:
        python_docs:
          type: "path"
          path: "docs/python/*.md"
          watch: true  # Update when files change

        code_standards:
          type: "text"
          content: "Our coding standards..."

        test_results:
          type: "cli"
          command: "pytest --json"
          refresh_interval: 300  # seconds

        git_info:
          type: "callable"
          import_path: "myapp.git:get_info"

      # On-demand Prompts
      prompts:
        bug_analysis:
          type: "file"
          path: "prompts/bug_analysis.txt"

        code_review:
          type: "dynamic"
          import_path: "myapp.prompts:generate_review_prompt"

File-based Environment

You can also move the environment configuration to a separate file:

agents:
  code_analyzer:
    environment:
      type: "file"
      uri: "environments/code_analyzer.yml"

Components

Tools

• Python functions that agents can call
• Can be imported from any installed package
• Support settings and confirmation requirements
• Can be chained together by capable agents

Resources (On-demand)

Resources that are loaded only when requested by tools: - path: File patterns with optional watching - text: Inline text content - cli: Command output with refresh intervals - callable: Python function results - repository: Git repositories - source: Python source code

Prompts (On-demand)

Prompts that are rendered when needed: - file: Load from text files - dynamic: Generate using Python functions - template: Jinja2 templates with context - static: Inline prompt definitions

Difference to Knowledge Sources

• Environment: On-demand loading when tools need them

  environment:
    resources:
      docs:
        type: "path"
        path: "*.md"  # Loaded when a tool requests it
  

• Knowledge: Pre-loaded at agent startup

  knowledge:
    paths: ["docs/**/*.md"]  # Loaded when agent starts
  

Example With Separate File

environments/code_analyzer.yml:

tools:
  analyze_complexity:
    import_path: "myapp.tools:analyze_complexity"

resources:
  docs:
    type: "path"
    path: "*.md"

prompts:
  code_review:
    type: "file"
    path: "prompts/review.txt"