IOA Core is an open-source governance kernel for AI workflows.

It focuses on policy enforcement, evidence capture, immutable audit trails, memory-backed orchestration, and multi-model review patterns.

ioa-core v2.6.1 is the current stable public release, available on PyPI.

The core governance primitives (audit chains, evidence bundles, policy enforcement, memory fabric, multi-model quorum) are real and production-tested. Some advanced CLI surfaces and deeper docs continue to evolve — see docs/OSS_LAUNCH_READINESS_CHECKLIST.md for the current status.

• hash-chained audit logging
• evidence bundle generation
• policy and system-law framing
• memory fabric primitives
• offline and live provider smoke testing
• local examples for governed workflow and quorum-style review

For the current public feature boundary, see FEATURE_MATRIX.md.

pip install ioa-core

Then clone the examples and run them:

git clone https://github.com/orchintel/ioa-core.git
cd ioa-core

# Check the CLI entrypoint
ioa --help

# Scaffold a minimal project
python examples/00_bootstrap/boot_project.py /tmp/ioa-core-demo-project

# Run a governed workflow example
python examples/10_workflows/run_workflow.py

# Run an offline multi-model roundtable example
python examples/20_roundtable/roundtable_quorum.py "Analyze this code for security issues (ok)"

# Check environment health
python examples/30_doctor/doctor_check.py

# Smoke test the provider layer in offline mode
IOA_PROVIDER=mock python examples/40_providers/provider_smoketest.py

# Run the Ollama turbo-mode demo
python examples/50_ollama/turbo_mode_demo.py turbo_cloud

Examples run offline by default unless you explicitly enable live mode and set provider credentials.

For development (editable install with dev tools):

git clone https://github.com/orchintel/ioa-core.git
cd ioa-core
pip install -e ".[dev]"

Governed workflow example:

{
  "task": "Analyze code for security issues",
  "policy": "demo-governed",
  "result": "OK",
  "evidence_id": "ev-0001",
  "audit_chain_verified": true,
  "system_laws_applied": ["Law 1", "Law 5", "Law 7"]
}

Roundtable example:

{
  "quorum_approved": true,
  "approve_count": 3,
  "total_votes": 3,
  "evidence_id": "ev-rt-0001"
}

• immutable audit chain with hash continuity
• redaction support for sensitive values
• append-only JSONL logging with rotation and replay protection
• evidence bundle object for validations, metadata, and signatures

• system-law framing for governed execution
• policy hooks and validation paths
• support for audit-linked governance events

• multi-provider abstractions
• offline mock mode for repeatable examples
• provider smoke testing
• quorum-style review examples for multi-model workflows

• memory fabric package with hot and persistent stores
• SQLite, S3, and local JSONL backends
• encryption support for memory storage

• docs/examples/QUICKSTART.md
• docs/examples/WORKFLOWS.md
• docs/examples/ROUNDTABLE.md
• docs/examples/PROVIDERS.md
• docs/examples/OLLAMA.md
• docs/OSS_LAUNCH_READINESS_CHECKLIST.md

Live provider tests are optional and require real API keys.

export OPENAI_API_KEY=your-key
IOA_LIVE=1 IOA_PROVIDER=openai python examples/40_providers/provider_smoketest.py

If live keys are not configured, stay in offline mode and treat results as simulation/demo outputs rather than provider validation.

Before positioning IOA Core as a polished stable OSS product, the project still needs:

• aligned release metadata and version reporting
• removal of roadmap-style commands from deeper onboarding docs
• clean test collection and supported-version CI proof
• consistent model provenance rollout across evidence and audit-producing call sites
• clearer governance observability surfaces

Most AI orchestration stacks optimize for routing and output generation.

IOA Core is built around a different requirement: important AI workflows should also emit policy context, evidence, and auditable traces that can be inspected later.

That core substrate is intended to support higher-level OrchIntel products without forcing each downstream product to reinvent governance separately.

See CONTRIBUTING.md for development workflow and SECURITY.md for vulnerability reporting.