# NovaFabric

> Open-source CLI toolkit that captures, replays, diffs, and audits AI system runs.
> Local-first (data stays on your machine). No accounts, no telemetry. Python 3.11+.

NovaFabric is a time machine for AI systems. When an AI agent behaves differently
than it did yesterday — different outputs, a regression, an unexpected tool call —
NovaFabric lets you replay the original run, compare the two side by side, and trace
exactly what changed. It is built around five primitives: Asset Registry, Run Capsule,
Replay, Lineage, and Evidence Bundle.

Install with: pip install novafabric

## Core pages

- [Home](https://novafabric.dev/): overview, 90-second story, five primitives
- [Why now](https://novafabric.dev/why): motivation and non-goals
- [Concepts](https://novafabric.dev/concepts): the five primitives and four replay modes in depth
- [Install](https://novafabric.dev/install): three-command quickstart
- [Spec](https://novafabric.dev/spec): JSON Schema 2020-12 definitions for every format

## Interactive showcase

- [Run capsule inspector](https://novafabric.dev/showcase/capsule): inspect a real captured run in the browser
- [Lineage graph](https://novafabric.dev/showcase/lineage): interactive DAG — provenance, blast radius, replay chain
- [Replay & diff](https://novafabric.dev/showcase/replay): four replay modes with explicit guarantees
- [Asset registry](https://novafabric.dev/showcase/registry): eval-gated promotion for models, agents, prompts
- [Evidence bundle](https://novafabric.dev/showcase/evidence): signed in-toto DSSE Statement v1, verified in-browser

## Key facts

- Current version: v0.7.0 (local-first, open source, CLI-first)
- Four replay modes: exact, mocked, semantic, forensic
- Capsules stored in .novafabric/runs/ as schema-validated directories
- Evidence bundles use in-toto DSSE Statement v1 + ed25519 signatures
- Code repository: https://github.com/novafabric/novafabric
- Install: https://pypi.org/project/novafabric/

## Note

llms.txt is a community proposal (llmstxt.org). No major AI engine has confirmed
consuming it as of 2026-06. This file serves as a human-readable index and a
low-cost hedge should adoption increase.
