AIngram: Local-First Agent Memory
AIngram provides a revolutionary approach to agent memory management that seamlessly integrates retrieval signals to ensure the right context is found every time, all while operating locally without any cloud dependencies. The entire system lives within a single SQLite file, eliminating the need for external databases or APIs, while offering enhanced privacy and integrity features.

Key Features

• Multi-Signal Retrieval: AIngram runs three retrieval signals including full-text search (FTS5), semantic vector search, and knowledge graph traversal, combining them using Reciprocal Rank Fusion (RRF) for superior context retrieval.
• Local-Only Design: Everything is stored in one SQLite database file, allowing easy backups, sharing, and offline access. This makes it robust against infrastructure failures, ensuring durability.
• Cryptographic Integrity: Each entry in AIngram is Ed25519-signed and linked in a tamper-evident hash chain. This feature is essential for verifying data integrity, especially in shared environments.
• Knowledge Graph: Automatically extracts entities and relationships from memory entries, enabling advanced queries that traverse the memory graph, further enhancing retrieval contexts.

Performance Metrics

AIngram has been extensively benchmarked using the LongMemEval framework:

• Recall Accuracy: Achieved 95.5% recall in real-world noisy conversation histories for correct context retrieval.
• End-to-End Answer Accuracy: A complete pipeline accuracy of 72.8% across various question types using the gpt-4o-mini model.

Usage Example

AIngram provides a straightforward Python API for interacting with the memory store. Below is a sample code snippet for recalling important information:

from aingram import MemoryStore

with MemoryStore('./agent_memory.db') as mem:
    mem.remember('The API rate limit is 100 req/min. Exceeding it causes silent drops.')
    results = mem.recall('what do I need to know before deploying?', limit=5)
    for r in results:
        print(r.score, r.entry.content)

Why Choose AIngram?

Traditional memory systems often fall short due to their reliance on single retrieval signals. AIngram's hybrid approach allows it to excel, especially under precise query conditions, where exact terminology and entity relationships are essential.

Quick Installation

To get started with AIngram, simply install via pip:

pip install aingram

This will provide the core functionality for agent memory management.

Community and Support

For support and to connect with other users, join the community on Discord or visit the official website at aingram.dev.

AIngram is open-source under the Apache 2.0 license, making it accessible for personal, educational, and commercial use.