LedgerMind

v2.8.6 · Autonomous Memory Management System for AI Agents

LedgerMind is not a memory store — it is a living knowledge core that thinks,
heals itself, and evolves without human intervention.

Featured On

📈 Recent Activity

Last 14 days:

• 2,353 Git clones (647 unique cloners)
• Strong PyPI growth (hundreds of downloads in recent days)

Installed & used in:

• Gemini CLI (100% zero-touch, fully stable)
• Claude Code
• Cursor
• VS Code

Featured & Published:

• New article on Dev.to: "True Zero-Touch Autonomus Memory for AI Agents"
• Automaticaly accepted to PitchHut directory

✨ Why LedgerMind?

What is LedgerMind?

Most AI memory systems are passive stores: you write, you read, and if the
information becomes stale or contradictory — that is your problem. LedgerMind
takes a fundamentally different approach.

LedgerMind is an autonomous knowledge lifecycle manager. It combines a
hybrid storage engine (SQLite + Git) with a built-in reasoning layer that
continuously monitors knowledge health, detects conflicts, distills raw
experience into structured rules, and repairs itself — all in the background,
without any intervention from the developer or the agent.

Core Capabilities

Architecture at a Glance

Installation

# Basic install
pip install ledgermind

# With 4-bit vector search (recommended for CPU/Mobile)
pkg install clang cmake ninja
pip install llama-cpp-python
pip install ledgermind[vector]

Requirements: Python 3.10+, Git installed and configured in PATH.

Quick Start

Option A: Zero-Touch Automation (Recommended)

The easiest way to use LedgerMind is to install the LedgerMind Hooks Pack. This automatically configures your LLM client to retrieve context before every prompt and record every interaction without the agent needing to manually call MCP tools.

Client Compatibility Matrix

# Install hooks for your preferred client (vscode, claude, cursor, or gemini)
# Memory is installed in the parent directory (../.ledgermind) by default for best isolation.
ledgermind install gemini

Now, simply use your client as usual. LedgerMind operates entirely in the background.

Option B: Library (Direct Integration)

from ledgermind.core.api.bridge import IntegrationBridge

# NOTE: Using '.ledgermind' in the parent directory (outside the project root) 
# is the recommended standard. This keeps memory isolated from project code,
# prevents context pollution in analysis tools (like 'read_file'), and ensures
# memory is not accidentally committed to source control.
bridge = IntegrationBridge(
    memory_path="../.ledgermind", 
    vector_model="../.ledgermind/models/v5-small-text-matching-Q4_K_M.gguf"
)

# Inject relevant context into your agent's prompt
context = bridge.memory.search_decisions("database migrations", namespace="prod_agent")

# Record a structured decision with namespacing
bridge.memory.record_decision(
    title="Use Alembic for all database migrations",
    target="database_migrations",
    rationale="Alembic provides version-controlled, reversible migrations.",
    namespace="prod_agent"
)

Option B: MCP Server (Secure)

# Set your API key for security
export LEDGERMIND_API_KEY="your-secure-key"

# Start the MCP server
ledgermind-mcp run --path ledgermind

Key Workflows

Workflow 1: Multi-agent Namespacing — Isolation Within One Core

# Agent A decision
memory.record_decision(title="Use PostgreSQL", target="db", namespace="agent_a")

# Agent B decision (same target, different namespace)
memory.record_decision(title="Use MongoDB", target="db", namespace="agent_b")

# Search only returns what belongs to the agent
memory.search_decisions("db", namespace="agent_a") # -> Returns PostgreSQL

Workflow 2: Hybrid Search & Evidence Boost

LedgerMind uses Reciprocal Rank Fusion (RRF) to combine Keyword and Vector
search. Decisions with more "Evidence Links" (episodic events) receive a
+20% boost per link to their final relevance score.

Documentation

Benchmarks (February 26, 2Y, v2.8.6)

LedgerMind is optimized for high-speed operation on Android/Termux
as well as containerized environments. It includes built-in security for MCP and
REST endpoints.

Retrieval Performance (Jina v5 Small Q4_K_M)

License

LedgerMind is distributed under the Non-Commercial Source Available License
(NCSA).

LedgerMind — the foundation of AI autonomy.