What it does
Quivr V2 ingests content durably, makes text searchable within seconds, enriches it in the background and lets you follow a topic over time. The core stays generic, while file formats, AI models and business rules are added as plugins, so it can be adapted without forking the platform. Search can be lexical, semantic or hybrid. Text is keyword-searchable as soon as it is segmented, and embeddings arrive afterwards without blocking it. Saved queries and subscriptions turn newly searchable articles into matches, with keyword alerts, plain-language descriptions of what to watch for, and comparisons against stored embeddings to catch rephrased or translated articles. Connectors can pull from RSS and Atom feeds, Microsoft 365 mailboxes and X lists, and results are delivered by API, change feed or signed webhooks. It runs on PostgreSQL, S3-compatible storage, Temporal and Weaviate with locally served E5 embeddings, started with Docker Compose. The README says the project is at evaluation stage, the API is v0 and may change without notice, and it should not be run in production yet.
Capabilities
- Durable, idempotent ingestion that never silently drops content
- Lexical, semantic and hybrid search over the same corpus
- Alerts from keyword rules or plain-language descriptions
- Connectors for RSS and Atom feeds, Microsoft 365 mail and X lists
- Plugin system for formats, models and business rules
- REST API and CLI, with change feed and signed webhook delivery
When to use it
- You want to experiment with a self-hosted search and monitoring engine for news-like content streams
- You need alerts that catch rephrased or translated articles, not just exact keywords
- You are comfortable running an evaluation-stage project on your own infrastructure
