Skip to content

Token-Aware Context Compaction (CONCEPT:KG-2.1)

Overview

Intelligent context window management with three strategies (summarize_tools, drop_middle, progressive). Compaction summaries persist as EpisodeNode snapshots for cross-session context recall. Adapted from Goose's context_mgmt.

Implementation Details

  • Source Code: agent_utilities/knowledge_graph/memory/agent_context.py (ContextCompactor)
  • Pillar: KG

Documentation Coverage

This is an auto-generated dedicated concept page to ensure 100% documentation coverage across the ecosystem.

Cross-Session Chat Recall (CONCEPT:KG-2.1)

Overview

Keyword-based search across stored chat sessions using the KG Cypher backend. Adapted from Goose.

Implementation Details

  • Source Code: agent_utilities/knowledge_graph/retrieval/chat_search.py
  • Pillar: KG

Documentation Coverage

This is an auto-generated dedicated concept page to ensure 100% documentation coverage across the ecosystem.

Project-Aware Context (CONCEPT:KG-2.1)

Overview

Native support for Claude-style project rules. Backend automatically loads and injects AGENTS.md (Project Rules) into the system prompt for high-fidelity codebase awareness.

Implementation Details

  • Source Code: agent_utilities/knowledge_graph/core/agents_md.py
  • Pillar: KG

Documentation Coverage

This is an auto-generated dedicated concept page to ensure 100% documentation coverage across the ecosystem.

Elastic Context Operators (CONCEPT:KG-2.2)

Overview

5 atomic operators for elastic context orchestration: Skip, Compress, Rollback, Snippet, Delete (ContextOperator). Compress is expressively complete while specialized operators reduce hallucination risk. Includes checkpoint/rollback support for speculative context operations. Derived from LongSeeker (arXiv:2605.05191v1).

Implementation Details

  • Source Code: agent_utilities/knowledge_graph/memory/agent_context.py (ContextOperator, AgentContextManager)
  • Pillar: KG

Documentation Coverage

This is an auto-generated dedicated concept page to ensure 100% documentation coverage across the ecosystem.

Multi-Timescale Memory Dynamics (CONCEPT:KG-2.2)

Overview

Three-tier memory with timescale-aware exponential decay: Working (5min half-life), Episodic (4hr), Semantic (30-day). High-activation memories consolidate from Working→Episodic→Semantic via access-count thresholds. Relevance-weighted retrieval with keyword scoring. Derived from Continual Knowledge Updating (arXiv:2605.05097v1).

Implementation Details

  • Source Code: agent_utilities/knowledge_graph/memory/agent_context.py (TimescaleMemoryStore, MemoryTimescale)
  • Pillar: KG

Documentation Coverage

This is an auto-generated dedicated concept page to ensure 100% documentation coverage across the ecosystem.