Conversation
Introduce ExternalId as a first-class entity identity model with identityType and encoding fields, and expose it through SDK memory and graph types. Add public graph seeding APIs for entities, edges, facts, batch variants, entity resolution, and linking external IDs. These APIs let developers seed known identity graphs instead of relying only on fuzzy extraction. Prioritize deterministic external ID lookup during entity resolution before alias, lexical, fuzzy, vector, or description matching. Conflicting external IDs now fail loudly instead of silently reassigning identity. Persist external IDs in pgvector using a normalized typegraph_entity_external_ids table with scoped uniqueness, lookup indexes, and adapter methods for linking and resolving entity identifiers. Add deterministic graph query intent parsing with parser modes, strictness, parse metadata, and expanded query pattern coverage. This replaces answerSide-style intent with explicit source/target anchoring and strict versus soft graph retrieval behavior. Refine predicate normalization for co-founded relationships and update graph retrieval to use deterministic or LLM intent parsing according to caller configuration. Wire the new graph APIs through the Typegraph instance and cloud client, update public exports, bump the SDK package version, and add tests for query intent parsing, external ID resolution, seeded graph writes, conflict handling, and pgvector external ID normalization.
Remove passages as a managed graph storage/model concept. Chunks are now the only retrievable text unit, graph traversal returns chunk-backed results, and index ingestion no longer persists separate passage nodes. Replace passage-specific graph storage with canonical typed typegraph_graph_edges over entity, chunk, and memory endpoints. Keep typegraph_entity_chunk_mentions as raw mention evidence, leave legacy passage tables orphaned, and update pgvector deploy DDL/indexes for typed edges, chunk refs, fact keyword search, and exact chunk-ref filtering. Add QueryEntityScope with structured ExternalId support, OR semantics, filter/boost modes, graph-backed scope resolution, indexed chunk-ref prefiltering, direct fact/entity semantic and keyword search without traversal, and scoped graph traversal over typed chunk edges. Add entity-aware memory subjects and recall scoping without changing auth identity. remember/correct/conversation memory can resolve or upsert subject entities from deterministic external IDs, link memory nodes to entities through graph edges, constrain contradiction checks by entity context, and forget now invalidates memory graph associations. Expand tests for deploy DDL, typed graph-edge writes, passage removal, chunk graph retrieval, entity-scope chunk prefiltering, external-ID OR scope resolution, memory-only scoped recall without graph, unresolved external-ID recall, forget invalidation, and graph runner/query planner API changes.
Adds a centralized ontology registry for entity types, predicates, aliases, inverse direction, symmetry, prompt generation, and soft domain/range validation. Simplifies canonical predicates for B2B, enterprise, technical, and retained historical/narrative use cases while routing tense into metadata and alias cues into entity aliases instead of graph edges. Adds public graph entity maintenance APIs for mergeEntities and deleteEntity with invalidate/purge modes, plus pgvector transactional support for rewriting edges, facts, graph edges, mentions, and external IDs. Adds entity lifecycle columns, fact invalidation, active-entity filtering, and memory subject defaults for external-user flows. Updates extraction, query intent, predicate normalization, cloud client wiring, tests, and internal documentation to reflect the centralized ontology, artifact entity type, canonical predicates, alias semantics, temporal metadata, and entity health operations.
Rename the indexed/citable text carrier from document to source across the public SDK, query results, adapters, storage schema, events, tests, and docs. This makes document a normal ontology entity type and removes artifact from the graph vocabulary. Sources now own retrieval-facing fields like title, url, metadata, status, visibility, and subject; chunks expose sourceId; query results expose chunk.source. Add SourceSubject as a first-class ingestion declaration. A subject means the source is primary evidence for that entity, without requiring user-selected relation strings such as record_of/about/text_of. When graph storage is configured, ingest resolves or upserts the subject entity, creates PRIMARY_SOURCE_CHUNK edges to every chunk, writes source_subject mention evidence with confidence 1.0, and seeds extraction context with the subject. Update pgvector and sqlite storage from document_id/typegraph_documents to source_id/typegraph_sources, including source subject persistence and from_/to_ chunk-reference graph columns. Rename source CRUD/search adapter methods and update event names to index.source, source.update, and source.delete. BREAKING CHANGE: RawDocument, typegraphDocument, DocumentsApi, documentId, documentIds, documentFilter, document storage routes, and artifact ontology type have been removed. Use SourceInput, typegraphSource, SourcesApi, sourceId, sourceIds, sourceFilter, and document as an ontology entity type instead.
Treat null and undefined optional inputs as omitted across SDK, cloud, graph, memory, ingest, and adapter paths. Align memory APIs around primary input plus opts bags and add regression coverage for cloud/self-hosted parity.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.