Mengram

A second key feature group includes the Cognitive Profile and Multi-User Isolation. The Cognitive Profile is generated with a single get_profile() call and compiles the most relevant information from all three memory types into a ready-to-use system prompt. This eliminates the need for manual context curation and provides instant personalization for any LLM. Multi-User Isolation enables one API key to serve many end-users by simply passing a user_id parameter to each memory operation. Each user receives completely isolated semantic facts, episodic events, procedural workflows, and their own cognitive profile. This is critical for SaaS applications where each customer expects a private, personalized AI experience based solely on their own data. The isolation is enforced server-side, ensuring data security and compliance without extra configuration.

Third, Mengram offers advanced capabilities including Knowledge Graph, Smart Triggers, and Experience-Driven Procedures. The Knowledge Graph extracts entities and relations from conversations, building a structured graph that enables queries like "Ali works_at Uzum Bank" rather than vague text matches. Smart Triggers allow memory to proactively raise its hand — for example, reminding the AI about upcoming deadlines, detecting contradictions in user statements, or identifying workflow pattern changes. Experience-Driven Procedures automatically evolve procedural memory based on repeated successes or failures. If a procedure fails multiple times, it auto-generates a new version; if three similar successes occur, a new workflow is created. This self-improvement mechanism ensures that the AI's procedural knowledge becomes more accurate and efficient over time, without manual intervention.

The overall workflow is designed for simplicity and seamless integration. Developers can start by installing the mengram-ai Python or JavaScript package, or by configuring the MCP server for tools like Claude Desktop, Cursor, Continue, and Windsurf. Once connected, each agent interaction is captured: the conversation messages are passed to the add() method, which auto-extracts memories of all three types. On subsequent queries, the search() method retrieves relevant memories across all types, with optional reranking on higher pricing tiers. The ask() endpoint provides synthesized answers with citations, eliminating the need to wire up separate RAG. The get_profile() endpoint offers a compiled summary for instant personalization. This creates a continuous learning loop where each interaction improves the agent's understanding, making it smarter over time without any manual data management.

Concrete use cases demonstrate the transformative power of persistent memory. For coding assistants like Claude Code, Mengram remembers the developer's preferred stack, past fixes, and deployment procedures across sessions and even machine switches, so no context is lost and the assistant becomes increasingly helpful. In multi-agent systems using CrewAI, one agent can research a topic and another can act on that knowledge, with shared memory ensuring consistency across the crew. Voice agents built with Vapi or Retell can recognize returning callers and instantly recall their history, eliminating the need for callers to repeat themselves. Autonomous workflows for job applications or incident management can learn from past outcomes, adapting strategies autonomously. These scenarios show how memory turns a stateless AI into a personal, evolving assistant that grows with the user.

Target users include AI developers, indie developers, and production teams building personal AI assistants or agentic systems across various domains. Mengram integrates with major platforms and tools: it works out-of-the-box with Claude Desktop, Cursor, Codex, and Windsurf via MCP, and provides dedicated packages for LangChain and CrewAI. Pricing starts with a generous free tier offering 40 memory adds and 200 searches per month, perfect for experimentation. Paid plans begin at $5 per month for Starter, scaling to Business at $99 per month for high-volume needs, and Enterprise with custom options. All tiers include sub-users, webhooks, and team management. The product is open source under Apache 2.0, self-hostable, and supports 23 languages natively. In summary, Mengram provides the essential memory layer for AI agents, making them truly personal and adaptive without complex infrastructure.