Redleaf 2.5.1

Redleaf 2.5.1

Following the success of Redleaf 2.0, which brought expanded file formats, a smoother UI, and streamlined workflows, I am thrilled to detail the Redleaf 2.5 architecture, alongside our newest update: Patch 2.5.1.

Redleaf 2.5 completely transformed the project from a standalone research app into a foundational knowledge engine. Now, 2.5.1 gives that engine a massive cognitive upgrade.
🔥 What's New in 2.5.1: Hybrid Semantic Search for Agents

When Redleaf 2.5 introduced autonomous agents, they were limited to exact-keyword Full-Text Search (FTS). If an agent searched for "corporate theft" but your document used the word "embezzlement", the agent would hit a dead end and waste your GPU time guessing synonyms.

In patch 2.5.1, the agents get a semantic awakening.

The advanced agent API now utilizes True Hybrid Search, combining strict SQL filtering, exact-keyword FTS, and AI Vector Embeddings. By streaming vectors directly out of the database and using high-speed NumPy matrix multiplication, the API merges exact keyword matches with fuzzy conceptual matches using Reciprocal Rank Fusion (RRF).

The Result: Autonomous ReAct loops in Node Leaf are now drastically faster and smarter. Agents find highly dense, conceptually relevant facts on their very first try—bypassing the "synonym-guessing penalty" entirely to write vastly superior, deeply cited reports.
🏛️ The Redleaf 2.5 Foundation

If you missed the initial 2.5 release, here are the massive architectural upgrades that make the 2.5.1 semantic patch possible:

Overhauled AI Agent API & Node Leaf Integration
Redleaf 2.5 introduced a completely redesigned API built specifically for autonomous AI agents. Agents can programmatically traverse the Redleaf Knowledge Graph, execute searches, and pull primary source documents. This API overhaul provides direct, native integration with my newly co-developed companion app, Node Leaf. Node Leaf is a visual, node-based RAG studio and prompt orchestration engine that turns Redleaf into a playground for advanced AI research.

The Dual-Database Workflow
To support massive archives of 10,000+ documents without freezing the operating system or locking the web server, Redleaf uses a two-database processing pipeline. Heavy text extraction, spaCy NLP analysis, and vector embedding generation are routed through a high-throughput DuckDB staging workspace. Once the heavy lifting is done, the clean data is instantly "baked" down into a highly indexed SQLite database, ensuring the web dashboard and API responses remain lightning fast.

Virtual Folders and Cross-Platform Portability
Redleaf 2.5 also introduced virtual folders that link to directories outside the portable Redleaf instance. These are mapped via an .rlink alias, which acts as a path variable bridge. Combined with smart, case-insensitive path resolution for .rklf package imports, you can now seamlessly transfer entire knowledge bases between Linux, macOS, and Windows machines without losing document connections or metadata.

Redleaf 2.5

Redleaf 2.5

Following the success of Redleaf 2.0, which brought expanded file formats, a smoother UI, and streamlined workflows, I am thrilled to announce Redleaf 2.5. This update transforms Redleaf from a standalone research app into a foundational knowledge engine, featuring massive architectural upgrades and an entirely new way to interact with your data.

Overhauled AI Agent API & Node Leaf Integration
Redleaf 2.5 introduces a completely redesigned API built specifically for autonomous AI agents. Agents can now programmatically traverse the Redleaf Knowledge Graph, execute hybrid FTS searches, and pull primary source documents. This API overhaul provides direct, native integration with my newly co-developed companion app, Node Leaf. Node Leaf is a visual, node-based RAG studio and prompt orchestration engine that turns Redleaf into a playground for advanced AI research.

The Dual-Database Workflow
To support massive archives of 10,000+ documents without freezing the operating system or locking the web server, Redleaf 2.5 introduces a brand new, two-database processing pipeline. Heavy text extraction, spaCy NLP analysis, and vector embedding generation are now routed through a high-throughput DuckDB staging workspace. Once the heavy lifting is done, the clean data is instantly "baked" down into a highly indexed SQLite database, ensuring the web dashboard and API responses remain lightning fast.

Virtual Folders and Cross-Platform Portability
I have also introduced virtual folders that link to directories outside the portable Redleaf instance. These are mapped via an .rlink alias, which acts as a path variable bridge. Combined with smart, case-insensitive path resolution for .rklf package imports, you can now seamlessly transfer entire knowledge bases between Linux, macOS, and Windows machines without losing document connections or metadata.