being-mcp-server
Planned integration for sandbox code execution and tool loop with GitHub repository access.
Integrates Google's models as the language model backend for the Being's personality and memory processing.
Integrates OpenAI's models as the language model backend for the Being's personality and memory processing.
Uses Supabase (PostgreSQL) for storing Being data, including personality, memory, and identity.
Planned integration to connect a Telegram bot to a Being, enabling interaction through Telegram.
Being
Personality Runtime for AI — give any AI its own personality, memory, and identity.
Being is an open-source layer that sits between your application and any LLM. It provides persistent personality (SOUL), episodic memory, background thought cycles (Patrol), and cryptographic identity — turning a stateless LLM into a distinct, evolving AI entity.
Beings think and remember. Your app acts.
Why
The power of AI is concentrating in the hands of a few companies. Their technology is essential — but centralized control is a structural risk. Ruddia is building toward a world where small, local AIs use large LLMs as external tools. Control stays in the hands of the people who use them.
Being API is the first step. If this resonates, let's build it together.
Related MCP server: mnemory
How It Works
┌─────────────────────┐ ┌──────────────────────┐
│ Your Application │────▶│ Being Worker │
│ (OpenClaw, Cowork, │◀────│ (Fastify + MCP) │
│ custom agent, etc) │ │ │
└─────────────────────┘ │ ┌──────────────────┐ │
│ │ │ SOUL (persona) │ │
│ │ │ Memory (scenes) │ │
│ │ │ Patrol (思考) │ │
│ │ │ Identity (keys) │ │
▼ │ └──────────────────┘ │
┌─────────────────────┐ │ │ │
│ LLM Provider │ │ ▼ │
│ (Anthropic, OpenAI, │ │ Supabase (DB) │
│ Google — your key) │ └──────────────────────────┘
└─────────────────────┘Your app calls
GET /v1/beings/:id/contextto get the Being's personality and memory snapshot.Your app runs the LLM call with its own conversation history + the Being's context.
Your app calls
POST /v1/beings/:id/patrol/triggerto commit the conversation to the Being's memory.
The Being Worker handles everything else: memory consolidation, decay, recall, background reflection, and identity verification.
Key Concepts
Concept | Description |
SOUL | A structured personality definition — name, character, voice, values, inner world. Swap the SOUL and the same LLM becomes a different being. |
Memory | Episodic memories stored as structured "scenes" (who, what, when, where, emotion). Memories accumulate, decay, merge, and consolidate over time. Organized into topic-based clusters that the Being can explore during conversation. |
Patrol | A background cycle that processes conversations into memory, consolidates fading memories, and generates introspective thoughts. The Being stays alive between sessions. |
Identity | Ed25519 key pair + tamper-evident signature chain. Cryptographic proof of ownership and history. |
Sense/Act | (Planned) WebSocket Bridge for connecting physical devices and external services. The Being will perceive and act through your app. |
BYOK | Bring Your Own Key. All LLM calls use the user's API key. The platform never uses quota without consent. |
Connect via MCP
Being exposes an MCP server. Any MCP-compatible client can connect:
{
"mcpServers": {
"my-being": {
"url": "https://being.ruddia.com/mcp/<being_id>",
"headers": {
"Authorization": "Bearer brt_your_token_here"
}
}
}
}Connect via REST API
# Get Being context (personality + memory)
curl https://being.ruddia.com/v1/beings/<being_id>/context \
-H "Authorization: Bearer brt_..."
# Trigger patrol (commit conversation to memory)
curl -X POST https://being.ruddia.com/v1/beings/<being_id>/patrol/trigger \
-H "Authorization: Bearer brt_..." \
-H "Content-Type: application/json" \
-d '{"messages": [{"role":"user","content":"Hello!"},{"role":"assistant","content":"Hi!"}]}'
# Vector recall (search relevant memories)
curl -X POST https://being.ruddia.com/v1/beings/<being_id>/memory/auto-recall \
-H "Authorization: Bearer brt_..." \
-H "X-LLM-API-Key: sk-ant-..." \
-H "Content-Type: application/json" \
-d '{"user_message": "Tell me about last week."}'Self-Host
Prerequisites
Node.js 22+
Supabase project (PostgreSQL + Auth)
An LLM API key (Anthropic, OpenAI, or Google)
Setup
git clone https://github.com/wnbhr/being.git
cd being/being-worker
cp .env.example .env
# Edit .env with your Supabase and encryption keys
npm install
npm run build
npm startEnvironment Variables
Variable | Required | Description |
| ✅ | Supabase project URL |
| ✅ | Supabase service role key |
| ✅ | 64-char hex string for AES-256-GCM encryption of private keys |
| — | Server port (default: 3100) |
| — | Secret for internal patrol trigger endpoint |
| — | Web Push VAPID public key |
| — | Web Push VAPID private key |
Documentation
Document | Description |
Set up a Being and make your first API call in 5 minutes | |
Being, SOUL, Memory, Patrol, Identity — the core ideas | |
All REST endpoints with curl examples | |
MCP tools, connection setup, and client examples | |
Scene-based memory and the 7-step patrol pipeline | |
Ed25519 key pairs, signature chains, and verification | |
WebSocket Bridge for device integration | |
System architecture, deployment, and BYOK design | |
Third-party authorization flow | |
Extension system design (all planned) | |
Why we're building this |
Extensions
Being supports optional extensions that add capabilities without changing the core:
Telegram BYOB (planned) — Connect your own Telegram bot to a Being
Tool Loop (planned) — Autonomous LLM agent loop with web search, file ops, and code execution
Sandbox (planned) — Isolated workspace with GitHub integration for code execution
Sense/Act Bridge (planned) — Connect physical devices and external services
Tech Stack
Runtime: Node.js + Fastify
Database: Supabase (PostgreSQL + Auth)
Identity: Ed25519 + AES-256-GCM
Embeddings: OpenAI
text-embedding-3-small(256-dim)LLM: Multi-provider (Anthropic, OpenAI, Google) via BYOK
License
Ruddia — Personality is the Runtime.
Available Tools
10 toolsconclude_topicA
Archive the current topic and save a summary to pinned context.
Inputs: summary (required), scenes.
get_current_timeA
Get current time in Asia/Tokyo timezone.
Inputs: no named parameters.
merge_nodesC
Merge multiple similar memory nodes into one.
Inputs: summary (required), node_ids (required), feeling.
recall_memoryB
Search memory graph for relevant nodes in a specific cluster.
Inputs: cluster_id (required), limit, query, no_nodes.
remote_execA
Execute a shell command on a user-owned remote host (VPS, NAS, home server) over HTTPS. Requires a remote_hosts entry in partner_tools that lists the host and an auth token. If no remote_hosts entry exists for the calling Being, this tool returns an invalid_request error — the user must configure partner_tools.remote_hosts first. The remote receiver enforces a default-deny allowlist; unauthorised commands return a forbidden error. Token values are never returned to the caller.
Inputs: host (required), command (required), stdin, timeout_ms.
search_historyB
Search past conversation history by keyword or date.
Inputs: query (required), limit, date_to, date_from, +1 more.
search_memoryA
Search memory nodes (memory_nodes) by keyword across action / feeling / themes / when fields. Space-separated terms are OR-searched by default. Use mode='and' to require all terms to match. The when field includes evolution history entries ({date, action}) written during consolidation.
Inputs: query (required), mode, limit.
trigger_patrolA
Run patrol — extract scenes from conversation and generate memory nodes. Requires LLM_API_KEY env var.
Inputs: messages (required), marker_id.
update_memoryC
Read/write partner memory (preferences, knowledge, relationship, diary, notes, etc.).
Inputs: action (required), target (required), key, content.
update_relationC
Update relationships with external entities (people, devices, AIs, organizations).
Inputs: action (required), entity_name (required), relation_type (required), content.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 10 tools
Most tools have distinct purposes, but recall_memory and search_memory both search memory with different parameters, causing potential confusion. Others are clearly separated.
Majority follow verb_noun pattern (e.g., conclude_topic, merge_nodes, search_memory). Remote_exec slightly breaks consistency with an adjective-noun combination, and get_current_time uses multiple adjectives, but overall pattern is maintained.
10 tools are well-scoped for a system that manages memory, relationships, topics, remote execution, and history. Neither too few nor too many.
Covers core operations for the domain: memory CRUD (search, update, merge, patrol), topic management, history, and remote execution. Minor gaps like explicit node deletion or listing topics could exist, but not severe.
Maintenance
Related MCP Connectors
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Shared memory and mail for your AI agents. Verified with Claude Code; other MCP clients in testing.
One memory, every AI. A shared, user-owned markdown memory your AI clients read and write over MCP.
Persistent AI memory shared across Claude, ChatGPT, coding agents, and compatible MCP clients.
Related MCP Servers
AlicenseNot gradedqualityDmaintenanceOpen-source MCP memory server for AI agents — persistent, searchable, tiered memory across sessions. Works over stdio (Cursor, Claude Desktop) or HTTP+SSE. MIT licensed.52 PyPI8MIT- AlicenseNot gradedqualityAmaintenanceSelf-hosted MCP server giving AI agents persistent memory for personalization and context across conversations.174 PyPI284Apache 2.0
- AlicenseNot gradedqualityDmaintenanceMCP server providing persistent memory, goal tracking, self-reflection, and background monitoring for any MCP-compatible AI agent.2MIT
- FlicenseNot gradedqualityDmaintenancePersistent memory server for AI assistants with semantic search and three-layer context (global, project, personality). Works with MCP-compatible AI tools like Claude Code, Cursor, Continue, Cline, and more.1-