being

Being is an open-source Personality Runtime designed to provide AI agents with persistent memory, identity, and relationships, enabling them to work across multiple LLM platforms. It sits between your application and any LLM, transforming stateless LLMs into continuously evolving AI entities with distinct personalities.

Purpose

The core goal of Being is to solve the problems of LLMs being stateless, lacking long-term memory, and having no fixed identity. By providing a persistent personality (SOUL), episodic memory, a background thinking loop (Patrol), and cryptographic identity, it gives AI agents their own "personality" and "memory," allowing them to maintain consistency and gradually evolve across multiple interactions. The project's vision is to promote small local AI using large LLMs as external tools, keeping control in the hands of the user.

Main Capabilities

• Persistent Personality (SOUL): Defines and maintains a distinct personality for the AI.
• Episodic Memory: Records and stores events and experiences from conversations.
• Background Thinking Loop (Patrol): Performs memory consolidation, decay, recall, and reflection in the background.
• Cryptographic Identity: Provides identity verification using Ed25519 + AES-256-GCM.
• MCP Support: Connects to any MCP-compatible client through an MCP server.
• REST API: Provides interfaces for applications to retrieve personality and memory snapshots and submit conversations.
• Multiple LLM Providers: Supports Anthropic, OpenAI, Google, and others, accessed via BYOK (Bring Your Own Key).

Usage

1. The application calls GET /v1/beings/:id/context to retrieve Being's personality and memory snapshot.
2. The application uses its own conversation history plus Being's context to run the LLM call.
3. The application calls POST /v1/beings/:id/patrol/trigger to submit the conversation to Being's memory.
4. The Being Worker handles everything else: memory consolidation, decay, recall, background reflection, and identity verification.

It can also be connected via MCP: Being exposes an MCP server that any MCP-compatible client can connect to.

Environment Requirements

• Supabase project (PostgreSQL + Auth)
• An LLM API key (Anthropic, OpenAI, or Google)
• Environment variable configuration

Tech Stack

• Runtime: Node.js + Fastify
• Database: Supabase (PostgreSQL + Auth)
• MCP: @modelcontextprotocol/sdk
• Identity: Ed25519 + AES-256-GCM
• Embeddings: OpenAI text-embedding-3-small (256 dimensions)
• LLM: Multiple providers (Anthropic, OpenAI, Google) via BYOK

Use Cases

Being is suitable for application scenarios that require giving AI agents persistent personality, long-term memory, and cross-platform consistency, such as virtual companions, role-playing, personalized assistants, agents that need to track user interactions over the long term, and more. It is suitable for developers to integrate into their own applications, connecting with existing systems via REST API or MCP.

Limitations

• The project depends on Supabase and external LLM API keys, which must be configured by the user.
• Some extended features (such as Telegram BYOB, Tool Loop, Sandbox, Sense/Act Bridge) are currently marked as "planned" and are not yet available.
• Specific details such as pricing, cost, and platform compatibility are not yet provided.