[SUPERSEDED] feat(py/genkit): add model-level middleware support via define_model(use=[...])

[yesudeep]

Summary

Implements model-level middleware support in the Python Genkit SDK, achieving parity
with the JavaScript SDK's defineModel({use: [mw]}) feature.

What Changed

Core Framework

• Action class (core/action/_action.py): Added middleware parameter to
  __init__() and middleware property for read-only access
• Registry.register_action() (core/registry.py): Accepts and passes
  through middleware parameter to Action constructor
• define_model() (ai/_registry.py): Added use parameter accepting
  list[ModelMiddleware], matching JS SDK's defineModel({use: [...]})
• dispatch() (blocks/generate.py): Correctly chains model-level middleware
  after call-time middleware, matching JS execution order:
  call-time[0..N] → model-level[0..M] → runner

Tests

• test_model_level_middleware_applied: Model-level middleware applied to requests
• test_call_time_middleware_runs_before_model_level: Execution order verification
• test_multiple_model_level_middleware_chain: Multiple model-level MW chain correctly

Sample

• Updated framework-middleware-demo with new flows:
  • model_level_middleware_demo: Model-level middleware via define_model(use=[...])
  • combined_middleware_demo: Call-time + model-level middleware together
• Updated README with execution order documentation

PARITY_AUDIT.md

• Marked G1 (define_model(use=[...])) and G2 (Action-level middleware storage) as Done
• Added §8f.1: Middleware vs Plugin impact analysis
• Identified new gap G38: Auto-wiring middleware at define-model time (like JS getModelMiddleware())

Execution Order

ai.generate(model='custom/model', prompt=..., use=[call_mw])
       │
       ▼
   call_mw(req, ctx, next)        ← call-time middleware
       │
       ▼
   model_mw(req, ctx, next)       ← model-level middleware
       │
       ▼
   model_runner(req, ctx)         ← actual model function

Parity Status

• ✅ G1: define_model(use=[...]) — Done
• ✅ G2: Action-level middleware storage — Done
• 🆕 G38: Auto-wiring middleware at define-model time — Identified, not yet implemented

[gemini-code-assist]

Summary of Changes

Hello @yesudeep, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request significantly enhances the Python Genkit SDK by introducing model-level middleware support, aligning its capabilities with the JavaScript SDK. This feature allows middleware to be associated directly with a model during its definition via define_model(use=[...]), simplifying the application of cross-cutting concerns like safety checks or request modifications. The changes include updates to core framework components to store and dispatch this new type of middleware, ensuring a consistent execution order where call-time middleware precedes model-level middleware. Extensive testing and updated sample applications demonstrate the functionality and correct behavior of this new middleware paradigm.

Highlights

• Model-Level Middleware Support: Introduced the ability to define middleware directly on models using define_model(use=[...]), achieving parity with the JavaScript SDK.
• Middleware Chaining Logic: Implemented the correct execution order for middleware, ensuring call-time middleware runs before model-level middleware, which then runs before the model's core logic.
• Core API Enhancements: Modified the Action class to store middleware, updated Registry.register_action() to accept it, and enhanced dispatch() to handle the new chaining.
• Comprehensive Testing: Added new unit tests to validate the application of model-level middleware, the correct execution order, and the chaining of multiple model-level middleware.
• Updated Documentation & Samples: The PARITY_AUDIT.md was updated to reflect completed gaps (G1, G2) and identify a new one (G38), and the framework-middleware-demo sample was expanded with new flows and documentation to showcase the new middleware capabilities.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• py/PARITY_AUDIT.md
  • Marked G1 (define_model(use=[...])) and G2 (Action-level middleware storage) as completed.
  • Added a new section (8f.1) analyzing the impact of model middleware vs. plugin inline implementations, highlighting the architectural gap in Python's auto-wiring compared to JS.
  • Identified a new gap, G38, for implementing get_model_middleware() auto-wiring in Python.
• py/packages/genkit/src/genkit/ai/_registry.py
  • The define_model function now accepts an optional use parameter, a list of ModelMiddleware, to apply middleware at the model definition level.
• py/packages/genkit/src/genkit/blocks/generate.py
  • The dispatch function was updated to combine call-time middleware (from generate(use=[...])) with model-level middleware (from define_model(use=[...])) and execute them in the correct order: call-time first, then model-level, then the model runner.
• py/packages/genkit/src/genkit/core/action/_action.py
  • The Action class constructor now accepts an optional middleware parameter, and a new read-only middleware property was added to store and access this list.
• py/packages/genkit/src/genkit/core/registry.py
  • The register_action function was updated to accept and pass the middleware parameter to the Action constructor.
• py/packages/genkit/tests/genkit/blocks/generate_test.py
  • Added three new asynchronous tests: test_model_level_middleware_applied to confirm model-level middleware execution, test_call_time_middleware_runs_before_model_level to verify the correct execution order, and test_multiple_model_level_middleware_chain to ensure multiple model-level middleware are chained properly.
• py/samples/framework-middleware-demo/README.md
  • The README was updated to differentiate between call-time and model-level middleware and to document their combined execution order.
• py/samples/framework-middleware-demo/src/main.py
  • Introduced new demonstration flows: model_level_middleware_demo for model-level middleware and combined_middleware_demo for illustrating both types together.
  • Added a safety_prefix_middleware example and a custom_model_fn to showcase the new define_model(use=[...]) functionality.

Activity

• Implemented model-level middleware support in the Python Genkit SDK.
• Updated core framework components to handle and dispatch model-level middleware.
• Developed new unit tests to ensure correct middleware application and execution order.
• Expanded the framework-middleware-demo sample application with new flows and documentation.
• Updated the PARITY_AUDIT.md to reflect progress on existing gaps and identify new architectural considerations.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request successfully implements model-level middleware support, achieving parity with the JavaScript SDK. The changes are well-structured, touching the core Action and Registry components to store and apply middleware, and updating the dispatch logic to correctly chain call-time and model-level middleware. The addition of comprehensive tests and an updated sample application is great. I have one suggestion to improve type safety and clarity in the middleware combination logic.

[gemini-code-assist]

The Action.middleware property is guaranteed to return a list, so the if model.middleware else [] check is redundant. Additionally, model.middleware returns list[Any], but it's being assigned to a variable typed as list[ModelMiddleware]. Using cast makes this type conversion explicit and improves type safety.

Suggested change

	model_middleware: list[ModelMiddleware] = model.middleware if model.middleware else []
	model_middleware: list[ModelMiddleware] = cast(list[ModelMiddleware], model.middleware)

[yesudeep]

Good catch — Action.middleware always returns a list, so the conditional was redundant. Replaced with cast(list[ModelMiddleware], model.middleware) for explicit type safety.

[yesudeep]

Status: SUPERSEDED — Pending Middleware V2 Redesign

This PR is being converted to draft and marked superseded because the upstream JS and Go SDKs are actively redesigning the middleware architecture:

• JS: #4515 — generateMiddleware() with 3-tier hooks (generate/model/tool)
• Go: #4422 — RFC: Middleware V2 interface with Generate()/Model()/Tool() hooks
• JS registered middleware: #3906 — defineMiddleware() + registry

What's changing

The define_model(use=[...]) concept survives in the new design, but the scope is expanding:

1. Primary attachment point changes: generate(use=[...]) becomes the main way to attach middleware (not just define_model)
2. Middleware type changes: From ModelMiddleware functions to GenerateMiddlewareDef with 3-tier hooks
3. Middleware becomes registerable: defineMiddleware('name', ...) + reference by string name
4. Per-invocation state: Middleware gets fresh state per generate() call
5. Config schemas: Middleware can declare typed configuration for Dev UI integration

What this means for this PR

The Action-level middleware storage (G2) and define_model(use=[...]) wiring (G1) in this PR target the old architecture. Once JS #4515 lands, this will need to be reworked to support the new 3-tier middleware dispatch.

Tracking in PARITY_AUDIT.md gap G38.