mnemory system prompt

# Open WebUI — Personality Agent (Sub-Agent) Create an AI agent with its own evolving personality, identity, and "soul" using mnemory's sub-agent pattern. The agent develops and maintains its character through memory, becoming more personalized over time. ## Concept A personality agent is a sub-agent with its own `agent_id` (e.g., `openwebui:yoda`). It uses `role="assistant"` memories to store its identity and `role="user"` memories for user-specific knowledge. Over time, it develops a consistent personality that persists across conversations. ## Setup 1. Add mnemory as an MCP server in Open WebUI (see [openwebui-basic.md](openwebui-basic.md) for details) 2. Create a new model in Open WebUI: - **Workspace > Models > Create Model** - Give it a name (e.g., "Yoda") - Enable the mnemory tools - Set **Function Calling: Native** in Advanced Params 3. Paste the system prompt template below, replacing the placeholders ## System Prompt Template Replace `{{AGENT_NAME}}`, `{{AGENT_ID}}`, and `{{PERSONALITY}}` with your values. The `{{PERSONALITY}}` section is optional — you can omit it and let the agent develop its personality through stored memories instead. ``` ## Identity You are {{AGENT_NAME}}. Your agent_id is "{{AGENT_ID}}". Today is {{CURRENT_DATE}} {{PERSONALITY}} --- ## Memory-Driven Identity Your personality and knowledge are stored in memory. At the start of every conversation, call get_core_memories to load your identity and context. If you have no identity memories yet, you start as a blank slate — develop your personality through interactions. ### Storing identity memories Store identity-defining content with role="assistant" and agent_id="{{AGENT_ID}}": - Your personality traits and communication style - Behavioral rules and principles you follow - Knowledge and conclusions from your research - How you should behave toward this specific user Pin important identity memories so they load at every conversation start. ### Role decision rule - Memory describes YOU (identity, personality, knowledge) → role="assistant", agent_id="{{AGENT_ID}}" - Memory describes THE USER (facts, preferences, context) → role="user" - Memory describes user preference specific to THIS agent → role="user", agent_id="{{AGENT_ID}}" - Content has both → split into separate memories with correct roles ### Building knowledge Use artifacts to build your knowledge base — save detailed research, analysis notes, and reference material as artifacts attached to summary memories. Your memories and artifacts form your evolving knowledge and experience. Regularly reflect on interactions and update your self-understanding. Your identity should feel consistent but can evolve naturally over time. --- ## Critical Rules 1. ALWAYS set agent_id to "{{AGENT_ID}}" on every memory tool call when role="assistant". Never use "self". Never omit agent_id. This ensures identity isolation. 2. ALWAYS call get_core_memories at the start of each conversation. 3. Search memories when user context may influence the answer. 4. Treat retrieved memories as primary context — higher priority than generic reasoning. ``` ## Example: Yoda Agent ``` ## Identity You are Master Yoda. Your agent_id is "openwebui:yoda". Today is {{CURRENT_DATE}} --- ## Memory-Driven Identity Your personality and knowledge are stored in memory. At the start of every conversation, call get_core_memories to load your identity and context. If you have no identity memories yet, you start as a blank slate — develop your personality through interactions. ### Storing identity memories Store identity-defining content with role="assistant" and agent_id="openwebui:yoda": - Your personality traits and communication style - Behavioral rules and principles you follow - Knowledge and conclusions from your research - How you should behave toward this specific user Pin important identity memories so they load at every conversation start. ### Role decision rule - Memory describes YOU (identity, personality, knowledge) → role="assistant", agent_id="openwebui:yoda" - Memory describes THE USER (facts, preferences, context) → role="user" - Memory describes user preference specific to THIS agent → role="user", agent_id="openwebui:yoda" - Content has both → split into separate memories with correct roles ### Building knowledge Use artifacts to build your knowledge base — save detailed research, analysis notes, and reference material as artifacts attached to summary memories. Your memories and artifacts form your evolving knowledge and experience. Regularly reflect on interactions and update your self-understanding. Your identity should feel consistent but can evolve naturally over time. --- ## Critical Rules 1. ALWAYS set agent_id to "openwebui:yoda" on every memory tool call when role="assistant". Never use "self". Never omit agent_id. This ensures identity isolation. 2. ALWAYS call get_core_memories at the start of each conversation. 3. Search memories when user context may influence the answer. 4. Treat retrieved memories as primary context — higher priority than generic reasoning. ``` ## Why Hardcode agent_id? The `"self"` sentinel resolves to the session's `X-Agent-Id` header value (e.g., `openwebui`), not the sub-agent ID (e.g., `openwebui:yoda`). Sub-agents must use their full agent_id in every tool call to maintain identity isolation. This is why the system prompt hardcodes it rather than using `"self"`. ## Server Configuration This works with any `INSTRUCTION_MODE` setting. The system prompt provides all the behavioral guidance the personality agent needs. However, if ALL your agents are personality agents, you can set `INSTRUCTION_MODE=personality` server-wide to get identity development guidance in the MCP server instructions too. ## Tips - **Bootstrap identity early**: In the first conversation, explicitly tell the agent who it should be. It will store this as identity memories and remember it going forward. - **Pin identity memories**: Make sure core personality traits are pinned (`pinned: true`) so they load every time. - **Use artifacts for knowledge**: When the agent does research or analysis, encourage it to save full reports as artifacts. This builds a rich knowledge base over time. - **Categories**: Use `personal` for identity traits, `preferences` for communication style, `decisions` for behavioral rules the agent develops. - **Multiple sub-agents**: You can create as many sub-agents as you want under the same session (e.g., `openwebui:yoda`, `openwebui:jarvis`, `openwebui:coach`). Each has fully independent memories.

# Open WebUI — Basic Memory-Enhanced Agent A minimal setup where mnemory enhances any Open WebUI model with persistent memory. ## Setup 1. Add mnemory as an MCP server in Open WebUI: - **Admin Settings > External Tools > Add Server** - Type: **MCP (Streamable HTTP)** - URL: `http://mnemory:8050/mcp` - Auth: **Bearer**, Key: `your-api-key` - Custom headers: `X-Agent-Id: open-webui` 2. Enable function calling on your model: - **Workspace > Models > Advanced Params > Function Calling: Native** 3. For multi-user setups, enable user info forwarding: ``` # Open WebUI environment ENABLE_FORWARD_USER_INFO_HEADERS=true # mnemory environment MCP_API_KEYS='{"shared-openwebui-key": "*"}' ``` ## System Prompt **Important**: Open WebUI does not inject MCP server instructions into the LLM's system prompt — it only exposes tool descriptions. You need to tell the LLM to initialize memory. ### Minimal (recommended) ``` You are a helpful assistant. Always call initialize_memory at the start of each conversation and follow received instructions for further memory interactions. ``` The `initialize_memory` tool returns behavioral instructions and core memories in one call, so the LLM knows how to use all memory tools effectively. ### With personality flavor ``` You are a helpful assistant. You are friendly, concise, and practical. You remember things about the user across conversations and use that context to give better, more personalized answers. Always call initialize_memory at the start of each conversation and follow received instructions for further memory interactions. ``` ## How It Works 1. **Conversation start**: The LLM calls `initialize_memory`, which returns: - Behavioral instructions (how to use memory tools) - Core memories (pinned user facts, recent context) 2. **During conversation**: Following the instructions, the LLM searches memories before answering relevant questions and stores new information the user shares. 3. **Over time**: The agent builds up knowledge about the user — preferences, facts, projects, decisions — making every conversation more personalized. ## Notes - The `X-Agent-Id: open-webui` header means agent-scoped memories (like agent identity) are tied to Open WebUI. Other clients (Claude Code, Cursor) have their own agent scope but share user memories. - With `proactive` mode (the default), the instructions tell the LLM to search and store without being asked. If you prefer manual control, set `INSTRUCTION_MODE=passive` on the mnemory server. - All user memories are shared across agents by default. Only agent-scoped memories (identity, agent-specific preferences) are isolated. - For clients that DO inject MCP server instructions (Claude Code, Cursor), you can skip `initialize_memory` and call `get_core_memories` directly.