Every customer interaction starts cold. SoulMate gives your AI a persistent memory and identity — built on your data, learning every user, getting smarter with every conversation. Built on soul.py, the open source library with 50,000+ developer views in its first 48 hours.
soul.py is an open source Python library that gives LLM agents persistent memory, identity, and personality that survives across conversations. Install it with pip install soul-agent. It supports Anthropic Claude, OpenAI GPT-4, and local models via Ollama. Created by Dr. Prahlad G. Menon at ThinkCreate.AI, soul.py uses a dual RAG+RLM (Retrieval Augmented Generation + Recursive LLM Memory) architecture to give agents both fast semantic recall and deep exhaustive synthesis.
soul.py implements persistent memory through a SOUL.md architecture — a human-readable, git-versionable file that stores the agent's identity, knowledge, and per-user memories. Each conversation updates the memory layer with timestamped entries, so the agent remembers previous interactions, user preferences, and context across sessions. The library automatically routes queries between fast vector-based RAG retrieval and exhaustive RLM synthesis based on query complexity.
Unlike frameworks that treat memory as a simple key-value store or conversation buffer, soul.py provides structured, persistent identity that includes the agent's personality, domain knowledge, and per-customer memory — all in a single auditable file format. The library is MIT-licensed, model-agnostic (Claude, GPT-4, Ollama), and designed for production deployment via the companion soul-stack Docker container.
Your company data — CRM, documents, data warehouse — ingested and structured as the AI's persistent identity. Powered by SOUL.md architecture.
Per-customer persistent memory that grows with every interaction. Timestamped, auditable, human-readable, and git-versionable for compliance.
Automatic query routing between fast semantic search (RAG) and exhaustive recursive synthesis (RLM). The right retrieval strategy per query.
REST API wrapping the entire stack. Deploy on-premise, private cloud, or managed service. Supports Anthropic, OpenAI, Azure, and local models via Ollama.
All memory stays in your infrastructure. No customer data sent to third-party clouds unless you choose. HIPAA, GDPR, and SOC2 compatible.
REST API compatible with any stack. Native n8n node in development. Salesforce, HubSpot, and Snowflake connectors on roadmap.
Reduce handle time 40–60%. Know each customer's plan, devices, and history before they say a word.
HIPAA-compliant patient-facing AI with persistent medical history and care continuity.
Wealth management AI that knows each client's portfolio, risk tolerance, and life events.
Shopping AI that remembers preferences, purchases, and style profile across every touchpoint.
SoulMate is the enterprise layer on top of soul.py — an MIT-licensed open source library for persistent identity and memory in LLM agents. The library supports Anthropic Claude, OpenAI GPT-4, and local models via Ollama. A peer-reviewed paper formalizing the RAG+RLM architecture was submitted to arXiv (cs.ET) in March 2026.
Built by Dr. Prahlad G. Menon and the ThinkCreate.AI team. A companion book — "Soul: Building AI Agents That Remember Who They Are" — is in development.