Establish one canonical subject and follow it through every observable surface.

Documentation onboarding: compiler-first deployment for local, heterogeneous AI.

The reading path

2 min read

Start with motivation, end with the source.

Technical brief, architecture, ABI, PrismIR, repository, capability map.

Start with motivation. End with the source.

  1. Technical Brief — Why Prism exists, how it complements Bonsai, and the compiler/runtime boundary in one narrative.
  2. Architecture Overview — Follow one tensor from logical identity through representation search, physical planning, ComputeImage sealing, and runtime execution.
  3. ComputeImage ABI — Inspect the artifact boundary: logical tensors, physical tile layouts, execution views, payloads, plans, and receipts.
  4. PrismIR — See how semantic execution intent stays separate from target-specific lowering and heterogeneous physical planning.
  5. Repository — Explore the Rust ECS compiler/runtime, target backends, validation paths, tests, and current implementation boundary.
  6. Capability Map — See what is implemented today across replay, provenance, experiments, native runtimes, multimodal execution, and search.

Then inspect the edges

1 min read

After the first pass, inspect what the architecture does at the boundary.

Heterogeneous targets, capabilities, physics, roadmap.

After the first pass, inspect what the architecture does at the boundary.

  • Heterogeneous targets — CPU, Metal, ROCm/HIP, and XDNA/XDNA2 planning.
  • Current capabilities — Replay, provenance, experiments, runtimes, and media paths.
  • The Physics of Prism — The laws behind light, geometry, motion, and evidence.
  • Roadmap and boundary — What is validated, compile-verified, and still gated.

Claims

The reading path starts with motivation and ends with the source. The vocabulary is consistent across the path.

Compile-time architecture, numerical admission, target-aware planning, inspectable deployment artifacts, and evidence systems share one vocabulary across the docs.