thang.

RAGOps · v2.0.2 · OPEN SOURCE

Release AI changes with evidence.

Compare recorded RAG and agent traces against an accepted baseline. Review regressions before deciding to ship.

Python 3.11+ · Offline core · v2.0.2

  1. 1

    Recorded inputs

    Baseline and candidate traces from the same scenario.

    Scenario

    ragops-three-minute-demo

    Accepted baseline
    baseline.json

    Candidate traces
    candidate.json

  2. 2

    Compare behavior

    Deterministic evaluation of recorded behavior.

    Synthetic regression comparison
    MetricBaselineCandidate
    Citation coverage1.00.5
    Citation precision1.00.5
    Lexical groundedness1.00.6
  3. 3

    Apply policy

    Check gates against a versioned release policy.

    Release policy

    Contract 2.0

    Failed gates
    • candidate_release_gate
    • citation_coverage_regression
    • citation_precision_regression
    • groundedness_regression
  4. 4

    Review decision

    One canonical decision. Human owns final shipping.

    Release decision

    ⊘ BLOCK

    Evidence
    Synthetic demo
    Bundle verification
    verified
    Next step
    Review and address regressions

Synthetic regression demo. Expected canonical decision: BLOCK. Bundle integrity is verified; model safety is not certified.

Start locally

Python 3.11+. Run a disposable local demo.

See example output →
pipx install ragops==2.0.2
ragops demo --profile engineer --output ./demo-output
ragops evidence verify --bundle ./demo-output/evidence

Already have pipx? Use its isolated environment. Otherwise choose Python venv (macOS/Linux) or From source (macOS/Linux commands).

Try the regression demo

Recorded output from the source demo. This page does not run RAGOps or inspect your files.

The demo writes local reports to ./demo-output. Its BLOCK decision is intentional.

Read demo source →

Regression checks

Compare recorded traces for citation, grounding and other quality regressions.

Versioned policy

Keep gate thresholds and the canonical PASS/WARN/BLOCK decision inspectable.

Portable evidence

Export reports and verify evidence bundle integrity without network calls.

Reports & integrations

Export HTML, JSON, Markdown, JUnit, SARIF and GitHub Summary. API/workbench and vendor adapters are optional. The core does not make network calls. Evidence verification checks integrity, not model safety.

Read documentation →