Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.
Python agentagentic-aiagentic-frameworkagentic-workflowaiautomationclaudedeepseekgenaigptllmlow-codemcpnextjsno-codeopenaipythonskillsworkflow
"""Serializable prompt module; instruction ordering is deliberately unspecified."""
from pydantic import BaseModel, ConfigDict
from dify_agent.layers.prompt.configs import Config
from pydantic_ai import RunContext
from pydantic_ai.messages import UserContent
from pydantic_ai.toolsets import FunctionToolset
from dify_agent.runtime.context import Deps
class State(BaseModel):
model_config = ConfigDict(extra="forbid")
class Toolset(FunctionToolset[Deps]):
def __init__(self, name: str):
self.name = name
super().__init__(id=name, instructions=self._render_instructions)
def _render_instructions(self, ctx: RunContext[Deps]) -> str:
config = Config.model_validate(ctx.deps.layers[self.name]["config"])
return "\n\n".join([*_fragments(config.prefix), *_fragments(config.suffix)])
def build_user_content(self, deps: Deps) -> list[UserContent]:
return list(_fragments(Config.model_validate(deps.layers[self.name]["config"]).user))
def _fragments(value: list[str] | str) -> list[str]:
return [value] if isinstance(value, str) else value