AgentConfig

Base classes

Name	Children	Inherits
NodeConfig llmling_agent_config.nodes Configuration for a Node of the messaging system.	TeamConfig BaseACPAgentConfig AgentConfig	Schema

⋔ Inheritance diagram

graph TD
  94123207821136["agents.AgentConfig"]
  94123170552208["nodes.NodeConfig"]
  94123169468080["schema.Schema"]
  94123157077184["main.BaseModel"]
  139872072243680["builtins.object"]
  94123170552208 --> 94123207821136
  94123169468080 --> 94123170552208
  94123157077184 --> 94123169468080
  139872072243680 --> 94123157077184

🛈 DocStrings

Bases: NodeConfig

Configuration for a single agent in the system.

Defines an agent's complete configuration including its model, environment, and behavior settings.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/agent_configuration/

Source code in src/llmling_agent/models/agents.py

class AgentConfig(NodeConfig):
    """Configuration for a single agent in the system.

    Defines an agent's complete configuration including its model, environment,
    and behavior settings.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/agent_configuration/
    """

    model_config = ConfigDict(
        json_schema_extra={
            "x-icon": "octicon:hubot-16",
            "x-doc-title": "Agent Configuration",
        }
    )

    inherits: str | None = Field(default=None, title="Inheritance source")
    """Name of agent config to inherit from"""

    model: str | ModelName | AnyModelConfig | None = Field(
        default=None,
        examples=["openai:gpt-5-nano"],
        title="Model configuration or name",
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/model_configuration/"
        },
    )
    """The model to use for this agent. Can be either a simple model name
    string (e.g. 'openai:gpt-5') or a structured model definition.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/model_configuration/
    """

    tools: list[ToolConfig | str] = Field(
        default_factory=list,
        examples=[
            ["webbrowser:open", "builtins:print"],
            [
                {
                    "type": "import",
                    "import_path": "webbrowser:open",
                    "name": "web_browser",
                }
            ],
        ],
        title="Tool configurations",
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/tool_configuration/"
        },
    )
    """A list of tools to register with this agent.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/tool_configuration/
    """

    toolsets: list[ToolsetConfig] = Field(
        default_factory=list,
        examples=[
            [
                {
                    "type": "openapi",
                    "spec": "https://api.example.com/openapi.json",
                    "namespace": "api",
                },
                {
                    "type": "file_access",
                },
                {
                    "type": "composio",
                    "user_id": "user123@example.com",
                    "toolsets": ["github", "slack"],
                },
            ],
        ],
        title="Toolset configurations",
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/toolset_configuration/"
        },
    )
    """Toolset configurations for extensible tool collections.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/toolset_configuration/
    """

    session: str | SessionQuery | MemoryConfig | None = Field(
        default=None,
        examples=["main_session", "user_123"],
        title="Session configuration",
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/session_configuration/"
        },
    )
    """Session configuration for conversation recovery.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/session_configuration/
    """

    output_type: str | StructuredResponseConfig | None = Field(
        default=None,
        examples=["json_response", "code_output"],
        title="Response type",
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/response_configuration/"
        },
    )
    """Name of the response definition to use.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/response_configuration/
    """

    retries: int = Field(default=1, ge=0, examples=[1, 3], title="Model retries")
    """Number of retries for failed operations (maps to pydantic-ai's retries)"""

    output_retries: int | None = Field(default=None, examples=[1, 3], title="Output retries")
    """Max retries for result validation"""

    end_strategy: EndStrategy = Field(
        default="early",
        examples=["early", "exhaust"],
        title="Tool execution strategy",
    )
    """The strategy for handling multiple tool calls when a final result is found"""

    avatar: str | None = Field(
        default=None,
        examples=["https://example.com/avatar.png", "/assets/robot.jpg"],
        title="Avatar image",
    )
    """URL or path to agent's avatar image"""

    system_prompts: Sequence[str | PromptConfig] = Field(
        default_factory=list,
        title="System prompts",
        examples=[["You are an AI assistant."]],
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/system_prompts_configuration/"
        },
    )
    """System prompts for the agent. Can be strings or structured prompt configs.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/system_prompts_configuration/
    """

    # context_sources: list[ContextSource] = Field(default_factory=list)
    # """Initial context sources to load"""

    config_file_path: str | None = Field(
        default=None,
        examples=["/path/to/config.yml", "configs/agent.yaml"],
        title="Configuration file path",
    )
    """Config file path for resolving environment."""

    knowledge: Knowledge | None = Field(
        default=None,
        title="Knowledge sources",
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/knowledge_configuration/"
        },
    )
    """Knowledge sources for this agent.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/knowledge_configuration/
    """

    workers: list[WorkerConfig] = Field(
        default_factory=list,
        examples=[
            [{"type": "agent", "name": "web_agent", "reset_history_on_run": True}],
            [{"type": "team", "name": "analysis_team"}],
        ],
        title="Worker agents",
        json_schema_extra={
            "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/worker_configuration/"
        },
    )
    """Worker agents which will be available as tools.

    Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/worker_configuration/
    """

    requires_tool_confirmation: ToolConfirmationMode = Field(
        default="per_tool",
        examples=["always", "never", "per_tool"],
        title="Tool confirmation mode",
    )
    """How to handle tool confirmation:
    - "always": Always require confirmation for all tools
    - "never": Never require confirmation (ignore tool settings)
    - "per_tool": Use individual tool settings
    """

    debug: bool = Field(default=False, title="Debug mode")
    """Enable debug output for this agent."""

    usage_limits: UsageLimits | None = Field(default=None, title="Usage limits")
    """Usage limits for this agent."""

    tool_mode: ToolMode | None = Field(
        default=None,
        examples=["codemode"],
        title="Tool execution mode",
    )
    """Tool execution mode:
    - None: Default mode - tools are called directly
    - "codemode": Tools are wrapped in a Python execution environment
    """

    def is_structured(self) -> bool:
        """Check if this config defines a structured agent."""
        return self.output_type is not None

    @model_validator(mode="before")
    @classmethod
    def validate_output_type(cls, data: dict[str, Any]) -> dict[str, Any]:
        """Convert result type and apply its settings."""
        output_type = data.get("output_type")
        if isinstance(output_type, dict):
            # Extract response-specific settings
            tool_name = output_type.pop("result_tool_name", None)
            tool_description = output_type.pop("result_tool_description", None)
            retries = output_type.pop("output_retries", None)

            # Convert remaining dict to ResponseDefinition
            if "type" not in output_type["response_schema"]:
                output_type["response_schema"]["type"] = "inline"
            data["output_type"]["response_schema"] = InlineSchemaDef(**output_type)

            # Apply extracted settings to agent config
            if tool_name:
                data["result_tool_name"] = tool_name
            if tool_description:
                data["result_tool_description"] = tool_description
            if retries is not None:
                data["output_retries"] = retries

        return data

    @model_validator(mode="before")
    @classmethod
    def handle_model_types(cls, data: dict[str, Any]) -> dict[str, Any]:
        """Convert model inputs to appropriate format."""
        if isinstance((model := data.get("model")), str):
            data["model"] = {"type": "string", "identifier": model}
        return data

    def get_toolsets(self) -> list[ResourceProvider]:
        """Get all resource providers for this agent."""
        providers: list[ResourceProvider] = []

        # Add providers from toolsets
        for toolset_config in self.toolsets:
            try:
                provider = toolset_config.get_provider()
                providers.append(provider)
            except Exception as e:
                msg = "Failed to create provider for toolset"
                logger.exception(msg, toolset_config)
                raise ValueError(msg) from e

        return providers

    def get_tool_provider(self) -> ResourceProvider | None:
        """Get tool provider for this agent."""
        from llmling_agent.tools.base import Tool

        # Create provider for static tools
        if not self.tools:
            return None
        static_tools: list[Tool] = []
        for tool_config in self.tools:
            try:
                match tool_config:
                    case str():
                        if tool_config.startswith("crewai_tools"):
                            obj = import_class(tool_config)()
                            static_tools.append(Tool.from_crewai_tool(obj))
                        elif tool_config.startswith("langchain"):
                            obj = import_class(tool_config)()
                            static_tools.append(Tool.from_langchain_tool(obj))
                        else:
                            tool = Tool.from_callable(tool_config)
                            static_tools.append(tool)
                    case BaseToolConfig():
                        static_tools.append(tool_config.get_tool())
            except Exception:
                logger.exception("Failed to load tool", config=tool_config)
                continue

        return StaticResourceProvider(name="builtin", tools=static_tools)

    def get_session_config(self) -> MemoryConfig:
        """Get resolved memory configuration."""
        match self.session:
            case str() | UUID():
                return MemoryConfig(session=SessionQuery(name=str(self.session)))
            case SessionQuery():
                return MemoryConfig(session=self.session)
            case MemoryConfig():
                return self.session
            case None:
                return MemoryConfig()
            case _:
                msg = f"Invalid session configuration: {self.session}"
                raise ValueError(msg)

    def get_system_prompts(self) -> list[BasePrompt]:
        """Get all system prompts as BasePrompts."""
        from llmling_agent_config.system_prompts import (
            FilePromptConfig,
            FunctionPromptConfig,
            LibraryPromptConfig,
            StaticPromptConfig,
        )

        prompts: list[BasePrompt] = []
        for prompt in self.system_prompts:
            match prompt:
                case str():
                    # Convert string to StaticPrompt
                    static_prompt = StaticPrompt(
                        name="system",
                        description="System prompt",
                        messages=[PromptMessage(role="system", content=prompt)],
                    )
                    prompts.append(static_prompt)
                case StaticPromptConfig(content=content):
                    # Convert StaticPromptConfig to StaticPrompt
                    static_prompt = StaticPrompt(
                        name="system",
                        description="System prompt",
                        messages=[PromptMessage(role="system", content=content)],
                    )
                    prompts.append(static_prompt)
                case FilePromptConfig(path=path):
                    # Load and convert file-based prompt

                    template_path = Path(path)
                    if not template_path.is_absolute() and self.config_file_path:
                        base_path = Path(self.config_file_path).parent
                        template_path = base_path / path

                    template_content = template_path.read_text("utf-8")
                    # Create a template-based prompt
                    # (for now as StaticPrompt with placeholder)
                    static_prompt = StaticPrompt(
                        name="system",
                        description=f"File prompt: {path}",
                        messages=[PromptMessage(role="system", content=template_content)],
                    )
                    prompts.append(static_prompt)
                case LibraryPromptConfig(reference=reference):
                    # Create placeholder for library prompts (resolved by manifest)
                    msg = PromptMessage(role="system", content=f"[LIBRARY:{reference}]")
                    static_prompt = StaticPrompt(
                        name="system",
                        description=f"Library: {reference}",
                        messages=[msg],
                    )
                    prompts.append(static_prompt)
                case FunctionPromptConfig(arguments=arguments, function=function):
                    # Import and call the function to get prompt content
                    content = function(**arguments)
                    static_prompt = StaticPrompt(
                        name="system",
                        description=f"Function prompt: {function}",
                        messages=[PromptMessage(role="system", content=content)],
                    )
                    prompts.append(static_prompt)
                case BasePrompt():
                    prompts.append(prompt)
        return prompts

    def render_system_prompts(self, context: dict[str, Any] | None = None) -> list[str]:
        """Render system prompts with context."""
        from llmling_agent_config.system_prompts import (
            FilePromptConfig,
            FunctionPromptConfig,
            LibraryPromptConfig,
            StaticPromptConfig,
        )

        if not context:
            # Default context
            context = {"name": self.name, "id": 1, "model": self.model}

        rendered_prompts: list[str] = []
        for prompt in self.system_prompts:
            match prompt:
                case (str() as content) | StaticPromptConfig(content=content):
                    rendered_prompts.append(render_prompt(content, {"agent": context}))
                case FilePromptConfig(path=path, variables=variables):
                    # Load and render Jinja template from file

                    template_path = Path(path)
                    if not template_path.is_absolute() and self.config_file_path:
                        base_path = Path(self.config_file_path).parent
                        template_path = base_path / path

                    template_content = template_path.read_text("utf-8")
                    template_ctx = {"agent": context, **variables}
                    rendered_prompts.append(render_prompt(template_content, template_ctx))
                case LibraryPromptConfig(reference=reference):
                    # This will be handled by the manifest's get_agent method
                    # For now, just add a placeholder
                    rendered_prompts.append(f"[LIBRARY:{reference}]")
                case FunctionPromptConfig(function=function, arguments=arguments):
                    # Import and call the function to get prompt content
                    content = function(**arguments)
                    rendered_prompts.append(render_prompt(content, {"agent": context}))

        return rendered_prompts

avatar class-attribute instance-attribute

avatar: str | None = Field(
    default=None,
    examples=["https://example.com/avatar.png", "/assets/robot.jpg"],
    title="Avatar image",
)

URL or path to agent's avatar image

config_file_path class-attribute instance-attribute

config_file_path: str | None = Field(
    default=None,
    examples=["/path/to/config.yml", "configs/agent.yaml"],
    title="Configuration file path",
)

Config file path for resolving environment.

debug class-attribute instance-attribute

debug: bool = Field(default=False, title='Debug mode')

Enable debug output for this agent.

end_strategy class-attribute instance-attribute

end_strategy: EndStrategy = Field(
    default="early", examples=["early", "exhaust"], title="Tool execution strategy"
)

The strategy for handling multiple tool calls when a final result is found

inherits class-attribute instance-attribute

inherits: str | None = Field(default=None, title='Inheritance source')

Name of agent config to inherit from

knowledge class-attribute instance-attribute

knowledge: Knowledge | None = Field(
    default=None,
    title="Knowledge sources",
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/knowledge_configuration/"
    },
)

Knowledge sources for this agent.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/knowledge_configuration/

model class-attribute instance-attribute

model: str | ModelName | AnyModelConfig | None = Field(
    default=None,
    examples=["openai:gpt-5-nano"],
    title="Model configuration or name",
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/model_configuration/"
    },
)

The model to use for this agent. Can be either a simple model name string (e.g. 'openai:gpt-5') or a structured model definition.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/model_configuration/

output_retries class-attribute instance-attribute

output_retries: int | None = Field(default=None, examples=[1, 3], title='Output retries')

Max retries for result validation

output_type class-attribute instance-attribute

output_type: str | StructuredResponseConfig | None = Field(
    default=None,
    examples=["json_response", "code_output"],
    title="Response type",
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/response_configuration/"
    },
)

Name of the response definition to use.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/response_configuration/

requires_tool_confirmation class-attribute instance-attribute

requires_tool_confirmation: ToolConfirmationMode = Field(
    default="per_tool", examples=["always", "never", "per_tool"], title="Tool confirmation mode"
)

How to handle tool confirmation: - "always": Always require confirmation for all tools - "never": Never require confirmation (ignore tool settings) - "per_tool": Use individual tool settings

retries class-attribute instance-attribute

retries: int = Field(default=1, ge=0, examples=[1, 3], title='Model retries')

Number of retries for failed operations (maps to pydantic-ai's retries)

session class-attribute instance-attribute

session: str | SessionQuery | MemoryConfig | None = Field(
    default=None,
    examples=["main_session", "user_123"],
    title="Session configuration",
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/session_configuration/"
    },
)

Session configuration for conversation recovery.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/session_configuration/

system_prompts class-attribute instance-attribute

system_prompts: Sequence[str | PromptConfig] = Field(
    default_factory=list,
    title="System prompts",
    examples=[["You are an AI assistant."]],
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/system_prompts_configuration/"
    },
)

System prompts for the agent. Can be strings or structured prompt configs.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/system_prompts_configuration/

tool_mode class-attribute instance-attribute

tool_mode: ToolMode | None = Field(default=None, examples=["codemode"], title="Tool execution mode")

Tool execution mode: - None: Default mode - tools are called directly - "codemode": Tools are wrapped in a Python execution environment

tools class-attribute instance-attribute

tools: list[ToolConfig | str] = Field(
    default_factory=list,
    examples=[
        ["webbrowser:open", "builtins:print"],
        [{"type": "import", "import_path": "webbrowser:open", "name": "web_browser"}],
    ],
    title="Tool configurations",
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/tool_configuration/"
    },
)

A list of tools to register with this agent.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/tool_configuration/

toolsets class-attribute instance-attribute

toolsets: list[ToolsetConfig] = Field(
    default_factory=list,
    examples=[
        [
            {"type": "openapi", "spec": "https://api.example.com/openapi.json", "namespace": "api"},
            {"type": "file_access"},
            {"type": "composio", "user_id": "user123@example.com", "toolsets": ["github", "slack"]},
        ]
    ],
    title="Toolset configurations",
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/toolset_configuration/"
    },
)

Toolset configurations for extensible tool collections.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/toolset_configuration/

usage_limits class-attribute instance-attribute

usage_limits: UsageLimits | None = Field(default=None, title='Usage limits')

Usage limits for this agent.

workers class-attribute instance-attribute

workers: list[WorkerConfig] = Field(
    default_factory=list,
    examples=[
        [{"type": "agent", "name": "web_agent", "reset_history_on_run": True}],
        [{"type": "team", "name": "analysis_team"}],
    ],
    title="Worker agents",
    json_schema_extra={
        "documentation_url": "https://phil65.github.io/llmling-agent/YAML%20Configuration/worker_configuration/"
    },
)

Worker agents which will be available as tools.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/worker_configuration/

get_session_config

get_session_config() -> MemoryConfig

Get resolved memory configuration.

Source code in src/llmling_agent/models/agents.py

def get_session_config(self) -> MemoryConfig:
    """Get resolved memory configuration."""
    match self.session:
        case str() | UUID():
            return MemoryConfig(session=SessionQuery(name=str(self.session)))
        case SessionQuery():
            return MemoryConfig(session=self.session)
        case MemoryConfig():
            return self.session
        case None:
            return MemoryConfig()
        case _:
            msg = f"Invalid session configuration: {self.session}"
            raise ValueError(msg)

get_system_prompts

get_system_prompts() -> list[BasePrompt]

Get all system prompts as BasePrompts.

Source code in src/llmling_agent/models/agents.py

def get_system_prompts(self) -> list[BasePrompt]:
    """Get all system prompts as BasePrompts."""
    from llmling_agent_config.system_prompts import (
        FilePromptConfig,
        FunctionPromptConfig,
        LibraryPromptConfig,
        StaticPromptConfig,
    )

    prompts: list[BasePrompt] = []
    for prompt in self.system_prompts:
        match prompt:
            case str():
                # Convert string to StaticPrompt
                static_prompt = StaticPrompt(
                    name="system",
                    description="System prompt",
                    messages=[PromptMessage(role="system", content=prompt)],
                )
                prompts.append(static_prompt)
            case StaticPromptConfig(content=content):
                # Convert StaticPromptConfig to StaticPrompt
                static_prompt = StaticPrompt(
                    name="system",
                    description="System prompt",
                    messages=[PromptMessage(role="system", content=content)],
                )
                prompts.append(static_prompt)
            case FilePromptConfig(path=path):
                # Load and convert file-based prompt

                template_path = Path(path)
                if not template_path.is_absolute() and self.config_file_path:
                    base_path = Path(self.config_file_path).parent
                    template_path = base_path / path

                template_content = template_path.read_text("utf-8")
                # Create a template-based prompt
                # (for now as StaticPrompt with placeholder)
                static_prompt = StaticPrompt(
                    name="system",
                    description=f"File prompt: {path}",
                    messages=[PromptMessage(role="system", content=template_content)],
                )
                prompts.append(static_prompt)
            case LibraryPromptConfig(reference=reference):
                # Create placeholder for library prompts (resolved by manifest)
                msg = PromptMessage(role="system", content=f"[LIBRARY:{reference}]")
                static_prompt = StaticPrompt(
                    name="system",
                    description=f"Library: {reference}",
                    messages=[msg],
                )
                prompts.append(static_prompt)
            case FunctionPromptConfig(arguments=arguments, function=function):
                # Import and call the function to get prompt content
                content = function(**arguments)
                static_prompt = StaticPrompt(
                    name="system",
                    description=f"Function prompt: {function}",
                    messages=[PromptMessage(role="system", content=content)],
                )
                prompts.append(static_prompt)
            case BasePrompt():
                prompts.append(prompt)
    return prompts

get_tool_provider

get_tool_provider() -> ResourceProvider | None

Get tool provider for this agent.

Source code in src/llmling_agent/models/agents.py

def get_tool_provider(self) -> ResourceProvider | None:
    """Get tool provider for this agent."""
    from llmling_agent.tools.base import Tool

    # Create provider for static tools
    if not self.tools:
        return None
    static_tools: list[Tool] = []
    for tool_config in self.tools:
        try:
            match tool_config:
                case str():
                    if tool_config.startswith("crewai_tools"):
                        obj = import_class(tool_config)()
                        static_tools.append(Tool.from_crewai_tool(obj))
                    elif tool_config.startswith("langchain"):
                        obj = import_class(tool_config)()
                        static_tools.append(Tool.from_langchain_tool(obj))
                    else:
                        tool = Tool.from_callable(tool_config)
                        static_tools.append(tool)
                case BaseToolConfig():
                    static_tools.append(tool_config.get_tool())
        except Exception:
            logger.exception("Failed to load tool", config=tool_config)
            continue

    return StaticResourceProvider(name="builtin", tools=static_tools)

get_toolsets

get_toolsets() -> list[ResourceProvider]

Get all resource providers for this agent.

Source code in src/llmling_agent/models/agents.py

def get_toolsets(self) -> list[ResourceProvider]:
    """Get all resource providers for this agent."""
    providers: list[ResourceProvider] = []

    # Add providers from toolsets
    for toolset_config in self.toolsets:
        try:
            provider = toolset_config.get_provider()
            providers.append(provider)
        except Exception as e:
            msg = "Failed to create provider for toolset"
            logger.exception(msg, toolset_config)
            raise ValueError(msg) from e

    return providers

handle_model_types classmethod

handle_model_types(data: dict[str, Any]) -> dict[str, Any]

Convert model inputs to appropriate format.

Source code in src/llmling_agent/models/agents.py

@model_validator(mode="before")
@classmethod
def handle_model_types(cls, data: dict[str, Any]) -> dict[str, Any]:
    """Convert model inputs to appropriate format."""
    if isinstance((model := data.get("model")), str):
        data["model"] = {"type": "string", "identifier": model}
    return data

is_structured

is_structured() -> bool

Check if this config defines a structured agent.

Source code in src/llmling_agent/models/agents.py

def is_structured(self) -> bool:
    """Check if this config defines a structured agent."""
    return self.output_type is not None

render_system_prompts

render_system_prompts(context: dict[str, Any] | None = None) -> list[str]

Render system prompts with context.

Source code in src/llmling_agent/models/agents.py

def render_system_prompts(self, context: dict[str, Any] | None = None) -> list[str]:
    """Render system prompts with context."""
    from llmling_agent_config.system_prompts import (
        FilePromptConfig,
        FunctionPromptConfig,
        LibraryPromptConfig,
        StaticPromptConfig,
    )

    if not context:
        # Default context
        context = {"name": self.name, "id": 1, "model": self.model}

    rendered_prompts: list[str] = []
    for prompt in self.system_prompts:
        match prompt:
            case (str() as content) | StaticPromptConfig(content=content):
                rendered_prompts.append(render_prompt(content, {"agent": context}))
            case FilePromptConfig(path=path, variables=variables):
                # Load and render Jinja template from file

                template_path = Path(path)
                if not template_path.is_absolute() and self.config_file_path:
                    base_path = Path(self.config_file_path).parent
                    template_path = base_path / path

                template_content = template_path.read_text("utf-8")
                template_ctx = {"agent": context, **variables}
                rendered_prompts.append(render_prompt(template_content, template_ctx))
            case LibraryPromptConfig(reference=reference):
                # This will be handled by the manifest's get_agent method
                # For now, just add a placeholder
                rendered_prompts.append(f"[LIBRARY:{reference}]")
            case FunctionPromptConfig(function=function, arguments=arguments):
                # Import and call the function to get prompt content
                content = function(**arguments)
                rendered_prompts.append(render_prompt(content, {"agent": context}))

    return rendered_prompts

validate_output_type classmethod

validate_output_type(data: dict[str, Any]) -> dict[str, Any]

Convert result type and apply its settings.

Source code in src/llmling_agent/models/agents.py

@model_validator(mode="before")
@classmethod
def validate_output_type(cls, data: dict[str, Any]) -> dict[str, Any]:
    """Convert result type and apply its settings."""
    output_type = data.get("output_type")
    if isinstance(output_type, dict):
        # Extract response-specific settings
        tool_name = output_type.pop("result_tool_name", None)
        tool_description = output_type.pop("result_tool_description", None)
        retries = output_type.pop("output_retries", None)

        # Convert remaining dict to ResponseDefinition
        if "type" not in output_type["response_schema"]:
            output_type["response_schema"]["type"] = "inline"
        data["output_type"]["response_schema"] = InlineSchemaDef(**output_type)

        # Apply extracted settings to agent config
        if tool_name:
            data["result_tool_name"] = tool_name
        if tool_description:
            data["result_tool_description"] = tool_description
        if retries is not None:
            data["output_retries"] = retries

    return data

Show source on GitHub