Typesense Instant Search brings sub-second federated search, a universal command palette (Cmd+K / Ctrl+K), and optional AI-powered natural-language and hybrid vector search to CiviCRM, backed by the open-source Typesense engine.
It indexes contacts, contributions, financial transactions, relationships, events, participants, memberships, activities, cases, groups, tags, mailings, Afform pages, and SearchKit saved searches, while strictly enforcing CiviCRM permissions and record-level ACLs through API4 verification.
Key features
- Universal Command Palette (
Cmd+K/Ctrl+K): Instant typo-tolerant search across 15 built-in and custom API4 entities, with optional replacement for the native CiviCRM quicksearch box. - AI Natural-Language Search: Converts conversational queries into structured filters, date/amount ranges, and sort orders using Google Gemini, OpenAI, Azure OpenAI, Cloudflare Workers AI, or self-hosted vLLM models, with an inspectable interpretation banner in the UI.
- Hybrid Semantic & Vector Search: Combines lexical ranking with dense vector embeddings (local built-in ONNX
multilingual-e5-small, OpenAI, Gemini, or self-hosted TEI/vLLM), backed by a persistent content-addressed embedding cache to avoid redundant vectorization during rebuilds. - Declarative API4 Blueprints: Configure indexed fields, weights, infix matching, facets, and result presentation without writing PHP code.
- Zero-Downtime Operations: Asynchronous event-driven outbox synchronization and atomic collection alias switching with instant rollback.
Support the project
Typesense Instant Search is independently developed and currently has no institutional or corporate sponsor. If your organization benefits from this work, you can help fund its maintenance, documentation or future development.
To sponsor work on Typesense Instant Search or request professional support, please contact Jonathan Dahan.
License
Typesense Instant Search is distributed under the GNU Affero General Public License v3.0 or later.
