The deterministic compiler for AI-native data pipelines. You describe outcomes in plain English; Osiris compiles them into reproducible, production-ready manifests that run with the same behavior everywhere (local or cloud).

# Setup
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# Initialize configuration
osiris init

# Start conversation - describe what you want in plain English
osiris chat

• Compiler, not orchestrator - Others schedule what you hand-craft. Osiris generates, validates, and compiles pipelines from plain English.
• Determinism as a contract - Fingerprinted manifests guarantee reproducibility across environments.
• Conversational → executable - Describe intent; Osiris interrogates real systems and proposes a feasible plan.
• Run anywhere, same results - Transparent adapters deliver execution parity (local and E2B today).
• Boring by design - Predictable, explainable, portable — industrial-grade AI, not magical fragility.

Interactive HTML dashboard showing pipeline execution metrics and performance

Comprehensive run overview showing E2B cloud execution with <1% overhead

Detailed view of pipeline steps with row counts and execution times

$ osiris chat

You: "Identify customers inactive for 90 days and export to CSV for re-activation campaign"

Osiris: I'll help you identify inactive customers. Let me explore your database...
        Found tables: customers, orders, sessions. I'll create a pipeline that:
        1. Identifies customers with no activity in the last 90 days
        2. Exports their details for your re-activation campaign

        Here's the generated pipeline:
        [Shows OML pipeline with SQL queries and transformations]

        Shall I compile this for execution?

You: "Yes, compile and run it locally"

Osiris: ✓ Pipeline compiled (manifest hash: a3f2b1c4)
        ✓ Execution complete: 847 inactive customers exported to output/inactive_customers.csv
        View report: osiris logs html --open

• AI-native pipeline generation from plain English descriptions
• Deterministic compilation with fingerprinted, reproducible manifests
• Run anywhere with identical behavior (local or E2B cloud)
• Interactive HTML reports with comprehensive observability
• AI Operation Package (AIOP) for LLM-friendly debugging and analysis
• LLM-friendly with machine-readable documentation for AI assistants

Osiris provides machine-readable documentation for AI assistants:

• For Users: Share docs/user-guide/llms.txt with ChatGPT/Claude to generate pipelines
• For Developers: Use docs/developer-guide/llms.txt for AI-assisted development
• Pro Mode: Customize AI behavior with osiris dump-prompts --export and osiris chat --pro-mode

Run pipelines in isolated E2B sandboxes with <1% overhead:

# Run in cloud sandbox
osiris run pipeline.yaml --e2b

# With custom resources
osiris run pipeline.yaml --e2b --e2b-cpu 4 --e2b-mem 8

See the User Guide for complete E2B documentation.

Every pipeline run automatically generates a comprehensive AI Operation Package for LLM analysis:

# View AIOP export after any run
osiris logs aiop --last

# Generate human-readable summary
osiris logs aiop --last --format md

# Configure in osiris.yaml
aiop:
  enabled: true  # Auto-export after each run
  policy: core   # ≤300KB for LLM consumption

AIOP provides four semantic layers for AI understanding:

• Evidence Layer: Timestamped events, metrics, and artifacts
• Semantic Layer: DAG structure and component relationships
• Narrative Layer: Natural language descriptions with citations
• Metadata Layer: LLM primer and configuration

See AIOP Architecture for details.

For comprehensive documentation, visit the Documentation Hub:

• Quickstart - 10-minute setup guide
• User Guide - Complete usage documentation
• Architecture - Technical deep-dive with diagrams
• Developer Guide - Module patterns and LLM contracts
• Examples - Ready-to-use pipelines

• v0.2.0 ✅ - Conversational agent, deterministic compiler, E2B parity
• v0.3.0 ✅ - AI Operation Package (AIOP) for LLM-friendly debugging
• v0.3.1 ✅ - Fixed validation warnings for ADR-0020 compliant configs
• v0.3.5 ✅ - GraphQL extractor, DuckDB processor, test infrastructure improvements
• v0.4.0 (Current) ✅ - Filesystem Contract v1 - deterministic, reproducible directory structure
• M2 - Production workflows, approvals, orchestrator integration
• M3 - Streaming, parallelism, enterprise scale
• M4 - Iceberg tables, intelligent DWH agent

See docs/roadmap/ for details.

See CONTRIBUTING.md for development workflow, code quality standards, and commit guidelines.

Apache-2.0