Zero-Vector v3 acts like a sophisticated long-term brain for AI assistants and digital personas. In simple terms, it prevents an AI from "forgetting" important details once a conversation ends. By managing memories and personal traits over time, it allows AI agents to recognize recurring users, recall past preferences, and maintain a consistent personality. This makes interactions feel much more natural and personalized, as the AI can build on previous context rather than starting from scratch every time. Under the hood, the system uses a high-performance "hybrid" architecture that combines semantic vector search with knowledge graph technology. While standard vector databases find information based on general similarity, the knowledge graph component allows the system to map out specific relationships between people, concepts, and events. This dual approach means the AI doesn't just find a relevant memory; it understands how that memory connects to other data points, creating a structured web of intelligence that can be traversed and explored logically. For developers building advanced applications, the tool introduces powerful workflow orchestration through LangGraph integration. This enables "multi-agent" coordination where different AI components can work together on complex tasks, share a unified memory state, and follow multi-step reasoning chains. It even supports "human-in-the-loop" processing, allowing developers to set up approval gates where the AI pauses for human verification before proceeding. With 24 specialized tools and a production-ready infrastructure using PostgreSQL and Redis, it provides a robust foundation for creating highly intelligent, context-aware AI systems.
Category: AI & LLM Tooling
Tags: knowledge-graph, langgraph, postgresql, redis, vector-search
bash # Clone the repository git clone https://github.com/MushroomFleet/zero-vector-3.git cd zero-vector-3 # 1. Set up the Zero-Vector v2 server (Original System) cd zero-vector/server npm install npm run setup:database npm run generate:api-key # Note the generated key for the MCP config cp env.example .env # Add your OpenAI API key npm start # Runs on port 3000 # 2. Set up the Zero-Vector v3 server (LangGraph System) cd ../../zero-vector-3/server npm install cp env.example .env # Configure environment npm run setup:postgres # Requires PostgreSQL and Redis to be running npm run setup:infrastructure npm start # Runs on port 3001 # 3. Set up the MCP server cd ../../MCP npm install cp env.example .env # Edit .env with both server URLs and API keys npm start ---MCP/.env file first: Required Environment Variables (MCP/.env): - V2_SERVER_URL=http://localhost:3000 - V3_SERVER_URL=http://localhost:3001 - V2_API_KEY=your_generated_v2_key - V3_API_KEY=your_generated_v3_key Claude Desktop Configuration: Add this to your claude_desktop_config.json: json { "mcpServers": { "zero-vector-v3": { "command": "node", "args": [ "/path/to/zero-vector-3/MCP/dist/index.js" ], "env": { "V2_SERVER_URL": "http://localhost:3000", "V3_SERVER_URL": "http://localhost:3001", "V2_API_KEY": "your_v2_key", "V3_API_KEY": "your_v3_key" } } } } ---create_persona: Create a new AI persona. - list_personas: List all available personas. - get_persona: Retrieve detailed persona configuration. - update_persona: Modify persona settings. - delete_persona: Remove a persona. Memory Management: - add_memory: Store a new semantic memory. - search_persona_memories: Search via vector similarity. - get_full_memory: Retrieve full memory context. - add_conversation: Log conversation history. - get_conversation_history: Retrieve historical chat logs. - cleanup_persona_memories: Automated memory maintenance. Knowledge Graph Tools: - explore_knowledge_graph: Traverse relationship nodes. - hybrid_memory_search: Combined vector and graph search. - get_graph_context: Retrieve context via graph entities. - get_graph_stats: View knowledge graph metrics. Workflow Operations (v3.0): - execute_workflow: Start a LangGraph multi-agent workflow. - get_workflow_status: Check status of a running execution. - resume_workflow: Handle human-in-the-loop approval gates. - cancel_workflow: Terminate an active workflow. - list_active_workflows: View all current executions. - get_workflow_metrics: View performance and reasoning analytics. Utility: - get_system_health: Check v2/v3 server status. - get_persona_stats: View specific persona statistics. - test_connection: Validate MCP-to-Server connectivity. ---zero_vector_conversation workflow for the persona 'helpful_assistant'. My query is: 'Explain machine learning fundamentals for beginners.' Enable performance caching and a confidence threshold of 0.8." Multi-Step Reasoning with Approval: > "Run a multi_step_reasoning workflow as 'technical_expert' to compare NLP approaches for a chatbot. Ensure human approval is enabled for the final recommendation." Graph Exploration: > "Explore the knowledge graph for the current persona and show me all entities related to 'Machine Learning' with a traversal depth of 2." Hybrid Search: > "Perform a hybrid memory search for 'project architecture' to find both semantically similar memories and their related graph entities."add_memory and explore_knowledge_graph tools allow the tutor to recall exactly what the student learned three weeks ago and how it relates to today’s lesson. Example: A student learns about "Variable Scope." Zero-Vector extracts "Variable Scope" as an entity and links it to "Python Basics." Next week, when the student asks about "Global Variables," the AI uses get_graph_context to realize the student already understands scope, allowing it to bypass introductory explanations and provide a more advanced lesson.execute_workflow tool can be set to a multi_step_reasoning type with enable_approval: true. The AI performs the heavy lifting of data analysis, pauses its state, and waits for a human to review the findings via resume_workflow. Example: An AI auditor scans a codebase for vulnerabilities. It finds a potential SQL injection. Instead of just flagging it, it uses a workflow to trace the data flow. Before it suggests a patch, it sends the "reasoning chain" to the Lead Developer. The Developer reviews the logic in the MCP client, hits "Approve," and the AI then generates and implements the specific fix.Part of MCP Servers