oris-ai 0.5.0

Oris: Responsible AI runtime framework for production pipelines.

CI
Docs
Package
Meta

Oris is an open-source Responsible AI runtime for Python: define pipelines in YAML or in code, run them through a framework-agnostic executor, and keep safety, observability, and validation on by default.

Design sequential pipelines with pluggable components, provider backends, and guardrails on inputs and outputs. Use the CLI for validate/run flows, or embed Oris in your own services with typed APIs and run- and step-level tracing for debugging and governance.

Table of contents

• Why Oris
• Features
• Installation
• Quick start
• Documentation (web)
• Project layout
• Development quality gates
• Governance and standards
• License

Why Oris

• Responsible by default: pipeline runs pass through input and output guards.
• Framework-agnostic runtime: integrate with any object that exposes a run(...)-style entry point.
• Production-oriented architecture: typed interfaces, clear module boundaries, strict CI quality gates.
• Traceable execution: run-level and step-level traces for debugging and governance workflows.

Features

YAML or embedded pipelines
Describe components and wiring in YAML, or build pipelines in Python with the same registry and executor.

Guards and policy
Basic policy hooks for harmful or sensitive inputs and outputs, with audit logging and redaction of sensitive fields.

Extensible components
Registry-based components, provider abstraction for model backends, and a CLI for validate and run.

Observability
Structured traces across the run so you can reason about what executed and when.

Installation

For library and CLI use from PyPI:

pip install oris-ai

For contributing or running tests from a clone:

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -e ".[dev]"

Quick start

Save the following as pipeline.yaml (any path you prefer):

name: basic_pipeline
components:
  - type: passthrough
    name: normalize_input
  - type: template_response
    name: responder
    config:
      template: "AI answer placeholder for: {query}"

Python (from the same directory as the file):

from oris import Pipeline

pipeline = Pipeline.from_yaml("pipeline.yaml")

result = pipeline.run({
    "query": "What is AI?"
})

print(result.output)

CLI:

oris validate pipeline.yaml
oris run pipeline.yaml --input-json '{"query":"What is AI?"}'

Documentation (web)

The full site (architecture, public API, security, releases, contributing) is published with MkDocs Material: devstrikertech.github.io/oris.

Private repositories need a GitHub plan that includes GitHub Pages; in repo settings choose Pages → Build and deployment → GitHub Actions after the first successful deploy from prod.

Local preview:

pip install -e ".[dev]"
bash scripts/sync_doc_sources.sh && mkdocs serve

Project layout

oris/
├── src/oris/
│   ├── core/
│   ├── runtime/
│   ├── components/
│   ├── providers/
│   ├── rai/
│   ├── integrations/
│   ├── tracing/
│   ├── cli/
│   └── api/
├── tests/
├── docs/
└── .github/workflows/

Development quality gates

• ruff for linting and format checks
• mypy in strict mode
• pytest and pytest-cov with a minimum of 84% coverage
• pre-commit hooks for local checks

Governance and standards

• Engineering standards: ENGINEERING_STANDARDS.md
• Security policy: SECURITY.md
• Architecture notes: ARCHITECTURE.md
• Public API guarantees: PUBLIC_API.md
• Provider system design: PROVIDER_DESIGN.md
• Release process: RELEASE.md
• Contribution process: CONTRIBUTING.md

License

MIT — see LICENSE in the repository root.