Summa 2 migration

Summa 2 uses the SpaceFrontiers/summa repository and the summa namespace throughout its packages, clients, commands, configuration, metrics and UI. This is a breaking namespace release, not a migration of the original Summa 0.x engine’s indexes or API. Rebuild 0.x indexes from their source documents.

Compatibility and ownership

The core remains the sole owner of search, scoring and index encodings. Native, async-only and WASM adapters use the same renamed core. Search-index format versions, magic bytes, field ordinals and encoded segment layouts are unchanged from the immediately preceding 1.9.1 implementation. Renaming does not add a query-time translation layer, allocation, cache, scorer or writer. Query and merge cost models are unchanged; no performance improvement is claimed.

Upgrade servers, brokers and clients together. Protobuf message field numbers and types are unchanged, but fully qualified RPC service names now begin with summa (and summa.broker for broker control). Regenerate bindings from summa-proto/summa.proto; mixed namespace deployments are unsupported.

  • Rust crates and binaries use summa-*; Rust imports use summa_*.
  • Python uses summa-client-python / summa_client_python and summa-mal / summa_mal. The TypeScript client is summa-client-typescript; the browser package is summa-wasm.
  • Rename environment variables to SUMMA_*, monitoring series to summa_*, Kubernetes labels to summa.spacefrontiers.org/*, and deployment names and volume references to their summa counterparts before restarting services.
  • Browser databases, local storage keys and model artifact cache directories use the new namespace. Export needed local browser data before upgrading; the new UI does not discover databases under the previous namespace.
  • Model trace identifiers and training hash-domain identifiers use the new namespace. Start fresh training runs and regenerate receipts, trace files and content-addressed training artifacts; search-index compatibility does not imply training checkpoint compatibility.

The original Summa blog and its assets are retained in the GitHub Pages source. Historical articles describe their original implementation; current package and API documentation is maintained alongside the code. The website build reads its API, schema and operations guides directly from these maintained sources through scripts/build_website.py, keeping the published guides in sync.

Release and validation

The first namespace release is 2.0.0. Internal crate requirements must match the released version, and crates must be published in dependency order. Packaged server and broker sources must contain their protobuf inputs so registry consumers can build them outside the monorepo.

The summa-proto source crate owns those inputs. Server and broker build scripts copy its embedded schemas and shared mutation validation into private Cargo output directories before invoking their existing generators. This adds build-time file copies only, with no runtime parser or search changes.

Validate the renamed search stack with python3 scripts/check_search.py check and full, regenerate both clients, build and test WASM, and check package contents and documentation links. Compare unchanged binary index fixtures and the protobuf schema’s field definitions; record results and environmental limitations in the performance review.

This site uses Just the Docs, a documentation theme for Jekyll.