Repository documentation
Documentation
Browse the documentation website or open the conference attendee hub for slides, the film, interactive labs and an offline download pack. The presenter guide includes a two-hour running order.
User documentation
- Current features and release posture
- Progressive visual labs
- Rebuildable instructional videos
- Watch the introductory courses
- Native gRPC oracle declarations and qualification boundary
- Independent child repository lifecycle commands
Start with Getting started for the shortest runnable path, then use the Literate AI user guide as the task map.
For a presentation-length introduction, use the editable Literate AI overview deck. It explains the thesis, lifecycle, evidence model, source-promotion path, cost controls, and enterprise agent-system boundary for a technically sophisticated new audience.
The Literate AI 1.3.1 release briefing summarizes the patch scope, core concepts, lessons from the first release onward, release-goal policy, current evidence, ETA, and publication path.
The Literate AI evolution conference deck connects the project's goals, release-by-release evolution, current 1.4.0 development surface, deliberate limits, and the visual lab progression.
For the current manager and engineering narrative, use the reproducible manager and engineering overview package. Maintained presentations and documents are modeled inside the Component system; read documentation artifacts as Components for the document-pair capability, its ecosystem Flavors, and the terminal-Component rule.
For the complete test story, read the three test boundaries, clean major rebuilds, and project test receipts together.
For the literate-documentation contract, start with the reviewed project graph and documentation authority review.
For dependency and environment reconstruction, read the CycloneDX SBOM and dependency graph together with cache/publication boundaries and retained Cargo package provisioning and secondary-account retained tests.
For the current per-Component implementation boundary, read
Per-Component execution plans. It separates
the implemented Standard-bound CLI, source scheduler, cache publication, and
admission/receipt core from deferred link/package and root-integration coverage plus the
remaining sample and qualification migration. Current
writers use the schemas/v2 catalog; schemas/v1 is
the frozen compatibility catalog.
For work that must survive an agent session, start with the active work queue and its user-directed work loop. This is the durable bridge from user direction to implementation, evidence, changelog, and reusable project learning.
For repository governance, release-line lockdown, patch authority, pull-request classification, and safe branch collection, read the Release Policy. The configuration reference lists the per-project fields and their defaults.
- Installation and first run
- Public repository export
- Test-suite audit (October 2026)
- Moving from NVIDIA-dev/literate-ai
- Getting started
- The framework flow
- Canonical project layout
- Core concepts and layered Components
- Writing readable specifications
- Components, identities, and exact versioning
- Flavors and target profiles
- Concurrent target matrices
- Private cross-platform test matrix
- Native FreeBSD installed SDK qualification
- Source to specification
- Model groups, selectors, and generation
- Caches, packages, and publication
- Source trust and build security
- Configuration and CLI reference
- HTML repository explorer
- Samples and tutorials
- OVA, migration, and rollback
- Troubleshooting
- Glossary
Architecture and policy
Migration and roadmap
- Active work resumption queue
- Prefix-installed CLI self-update
- Operator adoption program
- 0.9.0 executable-authority program
- Framework score-improvement program
- OVA two-phase rebase
- Phase 1 release and rollback
- Phase 2 lifecycle bridge
- Phase 2 compatibility prerelease and rollback
- Phase 2 final cutover
- Phase 8 source-to-specification roadmap
- Component authoring and lock separation roadmap
- Original-source native SDK consumption
Completed programs live under docs/history/roadmap/ with historical
lifecycle headers. Their owning queue items remain in
active work.