lexigram-ai-memory

AI memory system for the Lexigram Framework — episodic, semantic, and working memory

Overview

Three-tier AI memory system for the Lexigram Framework. Provides working, episodic, and semantic memory with pluggable backends, automatic consolidation scheduling, token-aware context assembly, and multi-source retrieval — all wired through the DI container via MemoryModule. Zero-config usage starts with sensible defaults.

Full documentation: docs.lexigram.dev

Install

uv add lexigram-ai-memory
# Optional extras
uv add "lexigram-ai-memory[redis]"

Quick Start

from lexigram import Application
from lexigram.di.module import Module, module

from lexigram.ai.memory import MemoryModule
from lexigram.ai.memory.config import MemoryConfig

@module(imports=[
    MemoryModule.configure(
        MemoryConfig(default_backend="in_memory")
    )
])
class AppModule(Module):
    pass

app = Application(modules=[AppModule])
if __name__ == "__main__":
    app.run()

Configuration

Zero-config usage: Call MemoryModule.configure() with no arguments to use defaults.

Option 1 — YAML file

# application.yaml
ai_memory:
  default_backend: "vector"
  ttl_seconds: 2592000
  consolidation:
    enabled: true
    interval_seconds: 3600.0

Option 2 — Profiles + Environment Variables (recommended)

export LEX_AI_MEMORY__DEFAULT_BACKEND=vector
# Environment variables for each field

Option 3 — Python

from lexigram.ai.memory.config import MemoryConfig
from lexigram.ai.memory import MemoryModule

config = MemoryConfig(
    default_backend="vector",
)
MemoryModule.configure(config)

Config reference

Module Factory Methods

Key Features

• Three-tier memory: Working (context assembly), Episodic (conversation episodes), Semantic (entity facts)
• Pluggable backends: In-memory, Redis, SQLAlchemy, Qdrant/Chroma/PGVector
• Token budget allocation: Distributes available tokens across memory sources
• Automatic consolidation: Background scheduler promotes episodic to semantic storage
• Multi-source retrieval: Unified MemoryRetriever queries all tiers with relevance ranking
• Dynamic pruning: DynamicContextPruner fits context into hard token limits

Testing

async with Application.boot(modules=[MemoryModule.stub()]) as app:
    # your test code
    ...

Key Source Files