Oxiverse is a privacy-first ecosystem of products including a search engine (IntentForge), browser, and productivity tools, all built on source-available principles and Privacy-by-Design.

Is Oxiverse open source?

Yes, Oxiverse is source-available under the Oxiverse Community License (OCL) v1.0, ensuring transparency and privacy for all users.

Where is the official Oxiverse source code?

The primary development repository is hosted on Codeberg (codeberg.org/oxiverse). A mirror exists on GitHub (github.com/oxiverse-ecosystem), but all contributions should be directed to Codeberg to ensure development remains within privacy-focused infrastructure.

IntentForge is an autonomous discovery engine utilizing intent extraction and self-healing search technology to provide private, relevant results without user profiling or data retention.

IntentForge v2 Architecture

IntentForge v2 is a high-performance discovery engine built to prioritize user intent over simple keyword matching. This document outlines the technical architecture, component interactions, and the hybrid search strategy.

High-Level Overview

IntentForge operates as a distributed system with a Rust core orchestrating specialized microservices.

Loading Diagram...

Core Components

1. Axum API (`src/api/`)

The entry point for all requests. It handles:

Search Logic: Orchestrates local Meilisearch queries and remote meta-search fan-out.
Caching: Multi-tiered caching (In-memory L1 + Redis L2).
Streaming: Supports SSE (Server-Sent Events) for real-time result updates.

2. Intent Classifier (`src/intent_classifier/`)

Analyzes queries to determine the user's goal (Informational, Navigational, Transactional, etc.).

Vectorized Mapping: Uses ONNX embeddings to map queries to intent groups.
Adaptive Search: Adjusts the semantic/keyword ratio based on query confidence.

3. Meta-Search Aggregator (`src/meta_search/`)

Discovers fresh content in real-time by querying external providers.

Fan-out Strategy: Parallelizes requests to multiple providers (DuckDuckGo, Bing, Brave, etc.).
Anonymization: All outgoing meta-search traffic is routed through Tor (Snowflake) or Cloudflare Worker Proxies to prevent tracking and IP blocking by upstream providers. (User-to-server traffic remains direct for performance).

4. Hybrid Search Engine

Combines the precision of keyword search with the context-awareness of semantic search.

Semantic Ratio: Defaults to 0.7 (70% semantic, 30% keyword).
Binary Quantization: Enables 8x vector compression for sub-50ms query latency.

5. Self-Improvement Service (`src/self_improvement/`)

An autonomous loop that triggers background indexing when search quality is low.

Gap Analysis: Identifies queries with zero or low-relevance results.
Proactive Discovery: Automatically searches, crawls, and indexes new content to fill gaps.

Data Flow: The Search Lifecycle

Request: User submits a query.
Classification: Intent classifier detects the query type and extracts salient attributes.
Local Search: Hybrid search queries the Meilisearch index.
Meta Fallback: If local results are insufficient, the aggregator fans out to meta-search providers.
Ranking: Results from all sources are merged, deduplicated (using SimHash), and re-ranked using a cross-encoder model.
Enrichment: Discovered URLs are queued for the background crawler to ingest full content for future queries.

Microservices

Query Layer: Python/FastAPI service for semantic query expansion and ranking.
Trafilatura: Specialized text extraction service for clean content retrieval.
YouTube-Unified: Optimized Node.js service for video discovery.

Infrastructure

Tor Transport: Native Tor daemon + Snowflake 2.12.1 with SQS rendezvous for censorship circumvention.
Meilisearch: Optimized for hybrid search with vector storage.
Redis: Handles deduplication (Bloom filters), caching, and discovery queues.