Skip to content

HORMER Hierarchical Navigation

Status: stable | Score: 100% | Last Tested: 2026-07-18

HORMER (Hierarchical Organized RAG Memory & Extraction Routing) implements a structured cognitive navigation framework. It utilizes a dynamic directory hierarchy, reinforcement learning (RL) scoring policies, Textual Gradient Descent (TGD) optimization, and GRPO to dynamically locate and organize memory clusters.

Memory documents are indexed into a tree-like hierarchy. The navigation policy scores sub-trees to guide agents directly to relevant knowledge branches without needing to parse the entire database flatly. Auto-improvement is achieved through textual gradient feedback loops where semantic retrieval mismatches adjust node rankings.

  • src/memory/hormer/ (HORMER nodes, navigation policies, and GRPO estimators)
  • src/navigation/ (tree traversal and navigation controllers)
  • Dynamic Directory Hierarchy: Organizes text chunks in cognitive directories.
  • Navigation Policy with Smart Scoring: Evaluates relevancy weight of semantic nodes.
  • Textual Gradient Descent (TGD): Auto-tuning and self-optimizing system prompts and weightings based on search hits.
  • GRPO Optimization: Simplified reinforcement learning algorithm adjusting policy outputs.
  • Shell Commands (API + CLI): Interface to manually traverse and check HORMER tree structures.
  • Navigation-aware consolidation tests.
  • GRPO step update validation and scoring convergence tests.
  • Phase 2 Polish is frozen for Product 1.0 stability. HORMER is fully operational in production.

Run hierarchical memory navigation for high-precision retrieval:

Terminal window
# Perform intelligent directory traversal using HORMER agent
xavier navigation explore "Find all security policy notes"

HTTP Request:

Terminal window
curl -X POST "http://localhost:8006/v1/navigation/explore" \
-H "X-Xavier-Token: $XAVIER_TOKEN" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Find all security policy notes",
"max_depth": 3
}'