ledgermind Files

Release Notes — Agent Memory System v2.4.0

This release, codenamed "The UX & Trust Update," focuses on strengthening agent confidence in retrieved context and delivering a seamless terminal experience.

🚀 Key Highlights in v2.4.0

1. Refined Runner (PTY Injection)
2. Auto-Enter Injection: The PTY driver now automatically submits queries after context injection. This eliminates the "double-Enter" friction, allowing for a truly fluid interaction with CLI agents.
3. Visual Stability: Resolved TUI flickering and "black square" artifacts. The injection process is now visually transparent and clean.

4. TTY Noise Filtering: Improved handling of control characters, ensuring stability when running inside complex TUI environments.

5. Verified Knowledge Base & Deep Context

6. "Verified" Markers: Memory blocks are now explicitly tagged as [VERIFIED]. This instructs LLMs to trust the injected context immediately, reducing redundant file system lookups and hallucinations.

7. Full Content Injection: Instead of passing mere file paths, the Runner now injects the complete content of decisions—including Rationale and Consequences. This provides the agent with the "why" behind every piece of knowledge, not just the "what."

8. Reliability & Vector Search

9. Stable Vector Retrieval: Fixed isolation rule bugs and improved the reliability of semantic search when handling large-scale local datasets.
10. Keyless Provider Support: Enhanced compatibility with local embedding models (e.g., all-MiniLM-L6-v2), enabling zero-config semantic search without external API keys.

🏗 Strategic Evolution (Hybrid Model Transition)

Version 2.4.0 marks the formalization of our new repository and licensing structure:

• OSS vs. Enterprise Split: The project is now divided into two independent repositories.
  • OSS Repository: Focuses on standard implementation, correctness, and agent autonomy ("How memory works").
  • Enterprise Repository: Focuses on governance, advanced auditing, and capability-based access control ("Who controls the memory").
• Hybrid Model Rationale: This split ensures physical isolation of license-protected code from the open-source core, simplifying dependency management and release cycles for different user segments.

📈 Improvements Over Previous Versions (v2.0.0 – v2.3.0)

• From Retrieval to Reasoning: Moving beyond simple storage, v2.4.0 leverages the Epistemic Hypothesis Model (v5) for deeper knowledge reflection.
• ACID Integrity: Building on the transaction support introduced in v1.20.0, the Semantic Store now guarantees total data integrity even during concurrent agent operations.
• Zero-Config Onboarding: With the introduction of local vector search in v2.3.0 and keyless embeddings in v2.4.0, the barrier to entry for local-first memory is now at its lowest.

📦 Getting Started To upgrade all components to v2.4.0:

pip install --upgrade -e ./core -e ./mcp_server -e ./adapters -e ./runner

Agent Memory System — Engineering the foundation of AI autonomy.