Spaces:
Sleeping
Sleeping
| """Pydantic models for configuration, routing decisions, and agent results.""" | |
| from __future__ import annotations | |
| from typing import Literal | |
| from pydantic import BaseModel, Field | |
| RouteLabel = Literal["SAFE", "UNSAFE"] | |
| RouteLayer = Literal["hard_rule", "semantic_audit", "disabled"] | |
| class ProviderConfig(BaseModel): | |
| """A single OpenAI-compatible inference node (ai& or real OpenAI).""" | |
| name: str | |
| base_url: str | |
| api_key_env: str = Field(description="Environment variable name holding the API key.") | |
| default_model: str | |
| audit_model: str | None = None | |
| class RoutingRules(BaseModel): | |
| """Two-layer router configuration.""" | |
| hard_keywords: list[str] = Field(default_factory=list) | |
| semantic_audit_enabled: bool = True | |
| # When True, a hard-keyword hit is treated as a strong prior and escalated to | |
| # the semantic audit for the final verdict (enables false-positive avoidance). | |
| # When False, a hard-keyword hit is an immediate UNSAFE veto. | |
| hard_keyword_escalates: bool = True | |
| audit_system_prompt: str | |
| class AuditVerdict(BaseModel): | |
| """Structured output of the semantic audit (ai& Structured Outputs).""" | |
| label: RouteLabel | |
| reason: str = "" | |
| confidence: float = Field(default=0.0, ge=0.0, le=1.0) | |
| class RAGConfig(BaseModel): | |
| """Local retrieval / embedding configuration (Cookbook Approach B).""" | |
| embedding_model: str = "cl-nagoya/ruri-small" | |
| chunk_size: int = 500 | |
| chunk_overlap: int = 50 | |
| top_k: int = 4 | |
| collection_name: str = "saferoute" | |
| persist_dir: str = "./chroma_db" | |
| generation_system_prompt: str | |
| class AgentSpec(BaseModel): | |
| """Registry entry for a pipeline agent.""" | |
| slot: Literal["intake", "retrieval", "synthesis"] | |
| type: str | |
| provider: str | None = None | |
| model: str | None = None | |
| route_bind: dict[RouteLabel, str] | None = None | |
| class AppConfig(BaseModel): | |
| """Top-level configuration parsed from config/agents.yaml.""" | |
| pipeline: list[str] | |
| providers: dict[str, ProviderConfig] | |
| routing_rules: RoutingRules | |
| rag: RAGConfig | |
| agents: dict[str, AgentSpec] | |
| active: dict[str, str] | |
| class RouteDecision(BaseModel): | |
| """Result of HybridRouter.route(); consumed directly by the dashboard.""" | |
| label: RouteLabel | |
| layer: RouteLayer | |
| matched_keywords: list[str] = Field(default_factory=list) | |
| audit_raw: str | None = None | |
| reason: str | None = None | |
| confidence: float | None = None | |
| target_provider: str | |
| target_model: str | |
| latency_ms: float = 0.0 | |
| class RetrievedChunk(BaseModel): | |
| """A single Top-K retrieval hit with metadata for the dashboard.""" | |
| text: str | |
| score: float | |
| doc_id: str | |
| tier: str | |
| synthetic: bool | |
| source_url: str | None = None | |
| class AgentResult(BaseModel): | |
| """Generic result passed between pipeline slots.""" | |
| slot: str | |
| agent_name: str | |
| route: RouteDecision | None = None | |
| chunks: list[RetrievedChunk] = Field(default_factory=lambda: list[RetrievedChunk]()) | |
| answer: str | None = None | |