Enterprise Edition

Release Notes

SWIRL 5, Enterprise Edition

SWIRL 5.0, released July 2026, is the current major release. Highlights:

Semantic Cache (separately licensed; license fields cache and cache_expiration): result caching, document version clustering, canonical version election, pinning, and freshness verification. See the Semantic Cache Guide.
Three-pass relevancy model: term and phrase match scoring, embedding-based semantic similarity, and cross-encoder re-ranking of the top results (enabled by an AI Provider with the reranker role). See the AI Search Guide.
Fetch more results: retrieve additional results on demand from every source that has more, without re-running the search.
Async, streaming search: results stream in as sources respond, with live per-stage progress (querying, waiting, re-ranking, fetching), token-by-token AI Insight streaming, and a Stop button for both search and AI Insight generation.
Topic Dashboard: a personal home-page dashboard of search topics with one-click folder caching, plus a configuration-driven Recent Files section for any connected OAuth2 source. See The Topic Dashboard.
Multi-model AI: AI Providers for all major LLM platforms via LiteLLM and direct connections, with per-provider models and prompts, streaming, and structured outputs. See the AI Search Guide.
Workspaces and members with owner/partner/editor/viewer roles and invites. See Workspaces and Members.
Built-in MCP server (swirl_mcp): stdio and streamable HTTP transports, 8 tools, read-only deployment resources. See the MCP Server guide.
Per-family prompt overrides and per-model RAG context budgets.
RAG answer persistence: AI Insights survive page reloads and are recorded for review.
Assistant upgrades: asynchronous chat with live progress, Markdown table rendering, answer feedback, and pinned-version awareness in RAG answers.
Duplicate handling: near-duplicate results are grouped rather than dropped, with a Show/Hide duplicates toggle; similarity detection uses a shared embedding space across all providers.
Result provenance: every result carries its source, the query as sent to that source, the fetch timestamp, and the response time.
Query transforms: configuration-driven query rewriting per SearchProvider, with CSV bulk-load of rewrite and synonym rules.
Search UX: grouped source selector, per-search notifications for infrastructure errors, and fast-fail on unreachable sources for quicker responses.
Keycloak (OIDC) login, with configurable brokered access to Microsoft 365 sources. See the Keycloak Guide.
Business Console: redesigned admin home with 30-day search and provider charts, Activity Analytics, a live Log Viewer, connector failure logging, create-by-pasting-JSON for any object, and white-label branding (page titles, logos, admin chrome).
OAuth refresh-token handling: connections to sources stay live across sessions.
Platform: Python 3.13, updated Apache Tika with OCR enabled by default, and in-process passage detection for RAG (no separate service required).
Docker-first deployment via Docker Compose. See the Installation Guide.

4.x and Earlier Releases

Documentation for 4.x deployments is archived under Enterprise 4.x Documentation.