Skip to content

Start here — bijux-vex onboarding

What problem does bijux-vex solve?

Vector execution often hides determinism assumptions. bijux-vex makes execution contracts explicit, auditable, and replayable so you can compare deterministic (exact) and non-deterministic (ANN) runs side by side.

When should you use it?

  • You need reproducible deterministic vector execution with provenance you can replay.
  • You want bounded, auditable non-deterministic execution (ANN) with explicit approximation and randomness metadata.
  • You care about comparing exact vs approximate results across backends and artifacts.

Deterministic vs non-deterministic (high level)

  • Deterministic: exact search, bit-stable, replay required. No hidden randomness.
  • Non-deterministic (ANN, experimental): approximate path only, bounded divergence, emits ApproximationReport + RandomnessProfile, replay is envelope-based.

Start here path (canonical)

1) CLI flow: ingest → materialize → execute → replay → compare. 2) Concepts: overview/concepts.md. 3) Example walkthrough: examples/overview.md.

Quickstart (Workflow A)

--8<-- "docs/examples/workflows/workflow_a.sh"

1) Concepts: overview/concepts.md — mental model of execution vs storage. 2) Contracts: spec/system_contract.md and spec/execution_contracts.md. 3) Examples: examples/overview.md — deterministic and ANN flows. 4) API: api/index.md and the canonical schema api/v1/schema.yaml. 5) Not a vector DB: user/not_a_vdb.md. 6) BYO vectors vs auto-embeddings: user/byo_vectors_vs_embeddings.md.