Content Intelligence Engine

Turn attributed source content into typed assessments that can be queried, exported and traced back to the original input.

The pipeline accepts bounded source events, applies assessment/relevance policies, validates backend answers and records decisions in a persistent ledger. It is designed for content analysis across news, websites and social/development sources; the package retains its omnia-news command names.

Source contract first

contracts.py requires source identity, platform, URL, author, text and observation time. It rejects unknown fields and oversized input, requires timezone-aware timestamps and checks source URLs. Supported platform values include github, x, reddit, youtube, website and news.

{
  "id": "documentation-event-1",
  "platform": "website",
  "url": "https://example.com/update",
  "author": "Example publisher",
  "text": "A synthetic release announcement.",
  "observed_at": "2026-10-20T12:00:00+00:00"
}

This is a synthetic source event, not an inference result. Context can retain bounded reply/quote/repost relationships; GitHub commit sources have additional repository/SHA identity checks.

Assessment lifecycle

flowchart TD
  Source[Attributed source event] --> Contract[Bounded input validation]
  Contract --> Policy[Task and relevance policy]
  Policy --> Backend[Configured assessment backend]
  Backend --> Answer[Typed-answer validation]
  Answer --> Ledger[(SQLite decision ledger)]
  Ledger --> Query[Query and JSONL export]

Loading

The pipeline, relevance policy and query interface are separate from the local inference adapter. Persistent records preserve evidence and policy context rather than retaining only a final label.

Run the offline checks

python -m venv .venv
python -m pip install -e .
python -m unittest discover -s tests -v

Tests use controlled backends and temporary state. The previously recorded clean installation/offline suite passed; it does not establish the quality of a particular checkpoint. New native inference requires the separately configured runtime and model revision. Use the package's CLI help and .env.example for runtime inputs; model weights and provider secrets are not bundled.

Technology: Python · typed contracts · local inference adapter · SQLite · JSONL · unittest

Source license: Apache-2.0. Existing upstream and third-party notices are retained.