rivet-ai 0.2.0

Add your description here

   ▄████████  ▄█   ▄█    █▄     ▄████████     ███    
  ███    ███ ███  ███    ███   ███    ███ ▀█████████▄
  ███    ███ ███▌ ███    ███   ███    █▀     ▀███▀▀██
 ▄███▄▄▄▄██▀ ███▌ ███    ███  ▄███▄▄▄         ███   ▀
▀▀███▀▀▀▀▀   ███▌ ███    ███ ▀▀███▀▀▀         ███    
▀███████████ ███  ███    ███   ███    █▄      ███    
  ███    ███ ███  ███    ███   ███    ███     ███    
  ███    ███ █▀    ▀██████▀    ██████████    ▄████▀  
  ███    ███                                         
  

Generate production-ready API clients from documentation in minutes, not days.

---

⚡ What is Rivet?

Rivet is not just a code generator; it is an autonomous software engineer dedicated to building Python SDKs.

Unlike standard OpenAPI generators that produce rigid, un-pythonic code, Rivet uses a multi-stage Agentic Architecture to read your Swagger/OpenAPI documentation, slice the specification to your exact needs, and generate robust, Pydantic-validated SDKs.

The Killer Feature: Rivet verifies its own work. It spins up a Docker container, runs pytest against the generated code, and if the tests fail, it reads the error logs and fixes the code automatically.

🚀 Key Features

🎯 Context Slicing: Don't need the whole 5,000-endpoint API? Rivet intelligently slices the OpenAPI spec to generate only the micro-SDK you requested (e.g., "Just the Payment endpoints").

🛡️ Type-Safety First: Generates strictly typed Pydantic v2 models for all requests and responses.

🔄 Self-Healing Loop: Generated code is executed in a sandbox. If pytest fails, the "Auditor" agent analyzes the stack trace and refactors the code until it passes.

📚 Documentation RAG: Scrapes your HTML documentation to enrich the SDK with better docstrings and edge-case handling than the raw spec provides.

🏗️ Architecture

Puff operates in a three-stage pipeline to ensure quality and correctness.

graph TD
    %% Node Definitions
    Ingest[("Ingest Node<br/>(Fetch Spec & Docs)")]
    Slice[("Slice Node<br/>(Filter Spec)")]
    GenSDK[("Generate SDK<br/>(LLM: Create client.py)")]
    ValSDK[("Validate SDK<br/>(Syntax & Import Check)")]
    GenTests[("Generate Tests<br/>(LLM: Create test_client.py)")]
    RunTests[("Run Tests<br/>(Pytest execution)")]
    
    FixSDK[("Fix SDK Targeted<br/>(LLM: Surgical Fix)")]
    FixTests[("Fix Tests Targeted<br/>(LLM: Adjust Tests)")]
    
    %% End Node
    End((End/Success))
    Fail((End/Fail))

    %% Styles with Darker Backgrounds and Lighter Text
    style Ingest fill:#023e58,stroke:#01579b,stroke-width:2px,color:#ffffff
    style Slice fill:#023e58,stroke:#01579b,stroke-width:2px,color:#ffffff
    style GenSDK fill:#1b5e20,stroke:#2e7d32,stroke-width:2px,color:#ffffff
    style ValSDK fill:#f57f17,stroke:#fbc02d,stroke-width:2px,color:#ffffff
    style GenTests fill:#1b5e20,stroke:#2e7d32,stroke-width:2px,color:#ffffff
    style RunTests fill:#f57f17,stroke:#fbc02d,stroke-width:2px,color:#ffffff
    style FixSDK fill:#bf360c,stroke:#d84315,stroke-width:2px,stroke-dasharray: 5 5,color:#ffffff
    style FixTests fill:#bf360c,stroke:#d84315,stroke-width:2px,stroke-dasharray: 5 5,color:#ffffff
    style End fill:#2e7d32,stroke:#1b5e20,stroke-width:2px,color:#ffffff
    style Fail fill:#c62828,stroke:#b71c1c,stroke-width:2px,color:#ffffff

    %% Main Flow
    Ingest --> Slice
    Slice --> GenSDK
    GenSDK --> ValSDK

    %% Validation Routing (route_after_sdk_validation)
    ValSDK -- "Status: sdk_valid" --> GenTests
    ValSDK -- "Status: sdk_invalid (Retry < 3)" --> FixSDK
    ValSDK -- "Status: sdk_invalid (Retry >= 3)" --> Fail

    %% SDK Fix Loop (Validation Cycle)
    FixSDK --> ValSDK

    %% Test Flow
    GenTests --> RunTests

    %% Test Routing (route_after_test)
    RunTests -- "Status: success" --> End
    RunTests -- "Status: test_failed" --> Analyze{Error Analysis}
    
    %% Error Analysis Routing
    Analyze -- "Is SDK Error? (Retry < 3)" --> FixSDK
    Analyze -- "Is SDK Error? (Retry >= 3)" --> Fail
    Analyze -- "Is Test Error? (Retry < 3)" --> FixTests
    Analyze -- "Is Test Error? (Retry >= 3)" --> Fail

    %% Test Fix Loop (Execution Cycle)
    %% Note: Test fixes go straight back to execution, skipping SDK validation
    FixTests --> RunTests

1. The Architect (Analysis)

Rivet pulls the full OpenAPI specification. If you asked for a specific feature set, the Spec Slicer traverses the reference graph to isolate only the relevant endpoints and their dependent data models, creating a "Mini-Spec."

2. The Builder (Generation)

A LangGraph Coding Agent, equipped with the Mini-Spec and RAG context from your documentation, generates the Python SDK (client.py) and a matching Test Suite (test_client.py). It adheres to strict PEP 8 and Pydantic v2 standards.

3. The Auditor (Verification)

This is where the magic happens.

1. The code is injected into an isolated Docker Container.

2. pytest is executed.

3. If tests pass: The SDK is packaged and delivered to you.

4. If tests fail: The Error Analyzer captures the stderr output, diagnoses the crash (e.g., ImportError, ValidationError), and feeds this feedback back to the Builder Agent to attempt a fix.

🛠️ Installation

# Clone the repository
git clone https://github.com/yashghogre/Rivet.git

# Install dependencies (We use uv!)
uv sync

💻 Usage

1. Run the Generator

You can generate a full SDK or scope it to specific requirements.

uv run rivet https://petstore.swagger.io/v2/swagger.json

2. Output

The final package will be available in the ./output directory:

output/
├── client.py            # The generated SDK
└── test_client.py       # The test suite used to verify the code