Axim Memory SDK

Axim Memory SDK™ is an enterprise-grade AI memory and extreme vector compression engine engineered by Cognitus Lab. Designed specifically for autonomous AI agents, multi-agent frameworks, and conversational pipelines, it provides instantaneous context recall and persistent local vector storage backed by a compiled, native Rust SIMD core.

⚠️ Proprietary Notice & License Terms

Axim Memory SDK™ is 100% Cognitus Lab Property. This software is licensed for execution only under the terms of the Cognitus Lab Proprietary End User License Agreement (EULA).

• Execution-Only License: Conditioned upon compliance with the EULA, users are licensed to download and run compiled packages to power AI workloads.
• Strict Restrictions: Resale, sublicensing, unauthorized mirroring, commercial redistribution, and decompilation/reverse-engineering are strictly prohibited.
• See the full LICENSE for legal terms.

Key Features

• ⚡ Native Rust SIMD Acceleration: Core quantization, bitwise Hamming distances, and vector projections are compiled directly into stripped native machine code via Maturin/PyO3.
• 🤖 Drop-in Agent Harness Connectors: First-class support for autonomous agent loops, prompt synthesis, and tool-calling schemas (OpenAI, Anthropic, Gemini, LangChain, CrewAI, AutoGen).
• 🛡️ Strict User & Tenant Isolation: Multi-tenant memory management ensures that queries and stored memories are strictly isolated by user_id.
• 📦 Zero External Runtime Python Dependencies: Pure standard library Python with a compiled native engine — zero third-party pip dependencies required.
• 💾 Crash-Resilient Persistence: Atomic disk-commit protocol guarantees zero file corruption during unexpected power loss or process termination.

Installation

Install the official package directly from PyPI:

pip install axim-memory-sdk

Agent Harness Integration (Quick Start)

Plug Axim Memory directly into any autonomous agent loop:

from axim_memory import AgentMemory

# 1. Define any vector embedding provider (OpenAI, Ollama, Cohere, etc.)
def get_embedding(text: str):
    return [0.1, -0.2, 0.4, 0.8] * 256  # 1024-dimension vector

# 2. Initialize the agent memory harness
memory = AgentMemory(
    embedding_fn=get_embedding,
    storage_path="~/.agent_memory.json",
    original_dim=1024,
    target_dim=128,
)

# 3. Record interactions and episodic facts
memory.remember_interaction(
    user_prompt="I prefer dark mode and concise TypeScript code",
    agent_response="Preferences noted: dark mode and TypeScript.",
    user_id="dev_user",
)

# 4. Inject memory context directly into your LLM system prompt
prompt_context = memory.build_system_prompt_context(
    query="Write a button component",
    user_id="dev_user",
)
print("System Prompt Context:\n", prompt_context)
# Output:
# <user_memory>
# - User: I prefer dark mode and concise TypeScript code Agent: Preferences noted...
# </user_memory>

# 5. Export as an Agent Tool Schema (OpenAI / Anthropic / Gemini function calling)
tool_schema = memory.as_tool()

Core Dual-Layer API (MultiLayerMemory)

For custom low-level retrieval architectures:

from axim_memory import MultiLayerMemory

mem = MultiLayerMemory(
    embedding_fn=get_embedding,
    original_dim=1024,
    target_dim=128,
)

# Store memory
mem.add("Project deadline is next Friday", user_id="alice")

# Search dual tiers (fast cache + persistent store)
results = mem.search("When is the project due?", user_id="alice")
print(results)

Support & Enterprise Licensing

For enterprise licensing inquiries, custom OEM runtime integration, or support:

• Organization: Cognitus Lab
• Repository: https://github.com/Cognitus-Lab/Axim-Memory-SDK