New features and improvements to Entity Enricher, newest first.
A new Semantic IDs page browses every concept your enrichments have resolved, shows how similar each one is to any concept you select, and flags near-duplicates sitting just below your threshold. Add concepts by hand in a keyboard loop that checks for an existing match before it creates anything, resolve a CSV of values in bulk, and explore a concept's neighbourhood as an orbit or a 3D map.
Semantic IDs →Switching embedding models used to be blocked once concepts existed. An owner can now migrate: every concept is re-embedded with the new model while enrichments keep running against the old one, and the switch happens automatically once coverage is complete. Concept IDs never change, so records, entities and database syncs stay valid — and a dry run shows the cost plus any concepts that would start converging.
Semantic IDs →Register a PostgreSQL, MySQL or SQLite database against one or more schemas, review the projected shape, then publish an additive DDL contract — every enrichment lands as constrained rows with real primary and foreign keys, per-schema delta webhooks, and replica migrations that keep the shape in sync as your schema evolves.
Database sync →Dates, times, UUIDs, emails, URIs, IP addresses and custom regex patterns are now first-class property attributes — schema generation infers them, validation enforces them, and database sync maps them to native column types.
Workflow Editor →The MCP server now ships with an embedded OAuth 2.1 authorization server — add Entity Enricher to claude.ai as a remote connector with no API keys to copy, and revoke access any time from the Connected Apps tab.
MCP data enrichment →Plans, limits, and FAQs are now published on a dedicated pricing page, hydrated from the live billing API and included in structured data for search engines.
Pricing →The whole product — marketing pages, documentation, and the signed-in app — is now available in 13 languages, including right-to-left Arabic, with locale-prefixed URLs for every public page.
Save a reusable benchmark scenario, run it against any set of models, and get automatic quality scores against a verified gold reference — completeness, correctness, and hallucination, side by side with cost and speed.
Model Benchmarks →Objects can now carry a stable, organization-scoped semantic ID, so the same real-world entity dedupes across enrichments, models, and languages via embedding similarity.
Semantic IDs →Entity Enricher now exposes a hosted Model Context Protocol server with 29 tools covering schema authoring, single and batch enrichment, fusion, and benchmarks — usable from Claude Desktop, Claude Code, and Cursor.
MCP Server →Attach PDFs, images, audio recordings, and office files to enrichment, schema generation, and playground requests — uploaded once, referenced by ID, delivered natively or as extracted text depending on the model.
Document Attachments →Expose a local Ollama to your hosted organization with one command — reverse WebSocket through the standard HTTPS ingress, per-tunnel revocable credentials, no public ports and no SSH.
Ollama Tunnels →