Skip to content

feat(integrations): add full cascade PydanticAI Model integration#179

Merged
saschabuehrle merged 1 commit intomainfrom
feat/pydantic-ai-integration
Apr 2, 2026
Merged

feat(integrations): add full cascade PydanticAI Model integration#179
saschabuehrle merged 1 commit intomainfrom
feat/pydantic-ai-integration

Conversation

@saschabuehrle
Copy link
Copy Markdown
Collaborator

Summary

  • Adds gold-standard PydanticAI integration with full cascade intelligence — matching the LangChain integration tier
  • CascadeFlowModel drops in as any PydanticAI Model, performing speculative drafting with quality gating, complexity-based pre-routing, domain policies, tool risk validation, and cost tracking with savings
  • Full harness integration (cost, latency, energy, budget gates) and streaming support

What's included

File Purpose
cascadeflow/integrations/pydantic_ai/model.py CascadeFlowModel — full cascade Model implementation
cascadeflow/integrations/pydantic_ai/config.py CascadeFlowPydanticAIConfig + DomainPolicy types
cascadeflow/integrations/pydantic_ai/quality.py Quality scoring bridge (core QualityValidator + heuristic fallback)
cascadeflow/integrations/pydantic_ai/harness_bridge.py Harness metrics recording + budget gate
cascadeflow/integrations/pydantic_ai/types.py CascadeResult, CostMetadata
cascadeflow/integrations/pydantic_ai/__init__.py Public API exports + create_cascade_model() factory
tests/test_pydantic_ai_integration.py 58 tests
docs-site/integrations/pydantic-ai.mdx Mintlify docs page
examples/integrations/pydantic_ai_harness.py Example usage
pyproject.toml Optional pydantic-ai extra

Usage

from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIModel
from cascadeflow.integrations.pydantic_ai import create_cascade_model

drafter = OpenAIModel("gpt-4o-mini")
verifier = OpenAIModel("gpt-4o")
cascade = create_cascade_model(drafter, verifier, quality_threshold=0.7)

agent = Agent(model=cascade)
result = await agent.run("What is quantum computing?")

Test plan

  • 58 PydanticAI integration tests pass
  • 114 other integration tests pass (OpenAI Agents, CrewAI, Google ADK)
  • 1022 core tests pass (1 skip for unrelated OpenAI rate limit)
  • Imports work with and without pydantic-ai installed
  • Integration registry shows pydantic_ai: True
  • CI green

Add gold-standard PydanticAI integration with full cascade intelligence:
speculative drafting with quality gating, complexity-based pre-routing,
domain policies, tool risk validation, cost tracking with savings
calculation, and harness metrics recording.

- CascadeFlowModel: drop-in PydanticAI Model with drafter→verifier cascade
- Pre-routing via ComplexityDetector for HARD/EXPERT queries
- Quality scoring bridge (core QualityValidator + heuristic fallback)
- Tool risk escalation for high-risk tool calls
- Domain policy overrides (threshold, force_verifier, direct_to_verifier)
- Streaming support with optimistic drafter + quality-gated escalation
- Full harness integration (cost, latency, energy, budget gates)
- 58 tests covering cascade flow, domain policies, tool risk, streaming,
  harness metrics, budget enforcement, and fail-open behavior
- Docs, README, example, and pyproject.toml optional dependency
@mintlify
Copy link
Copy Markdown

mintlify bot commented Apr 2, 2026

Preview deployment for your docs. Learn more about Mintlify Previews.

Project Status Preview Updated (UTC)
cascadeflow 🟢 Ready View Preview Apr 2, 2026, 6:35 AM

@saschabuehrle saschabuehrle merged commit c89e402 into main Apr 2, 2026
23 of 28 checks passed
@saschabuehrle saschabuehrle deleted the feat/pydantic-ai-integration branch April 2, 2026 09:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants