fleet-memory/hindsight-docs/src/data/templates.json

{
  "templates": [
    {
      "id": "conversation",
      "name": "Conversation",
      "description": "For chat-based agents and assistants. Tracks user preferences, conversation patterns, and builds a profile over time.",
      "category": "chat",
      "integrations": [
        "litellm",
        "langgraph",
        "pydantic-ai",
        "ai-sdk",
        "chat",
        "crewai",
        "ag2",
        "agno",
        "strands",
        "llamaindex",
        "local-mcp",
        "skills"
      ],
      "manifest": {
        "version": "1",
        "bank": {
          "retain_mission": "Extract user preferences, stated facts about themselves, requests they've made, topics they care about, and any commitments or follow-ups. Ignore small talk and filler.",
          "enable_observations": true,
          "observations_mission": "Track stable user preferences, communication style, recurring topics, and how the user's needs evolve over time."
        },
        "mental_models": [
          {
            "id": "user-profile",
            "name": "User Profile",
            "source_query": "What do we know about this user? What are their preferences, background, and how do they like to interact?",
            "max_tokens": 2048,
            "trigger": {
              "refresh_after_consolidation": true
            }
          },
          {
            "id": "open-threads",
            "name": "Open Threads",
            "source_query": "What topics, tasks, or follow-ups are still open or unresolved from past conversations?",
            "max_tokens": 1024,
            "trigger": {
              "refresh_after_consolidation": true
            }
          }
        ]
      }
    },
    {
      "id": "coding-agent",
      "name": "Coding Agent",
      "description": "For coding assistants. Remembers project architecture, technical decisions, coding patterns, and user preferences across sessions. High literalism for precise technical recall.",
      "category": "coding",
      "integrations": [
        "claude-code",
        "codex"
      ],
      "manifest": {
        "version": "1",
        "bank": {
          "retain_mission": "Extract technical decisions and their rationale, architectural choices, coding patterns and conventions, project structure facts, library/tool preferences, and recurring issues. Ignore transient debugging output and boilerplate.",
          "enable_observations": true,
          "observations_mission": "Track stable project facts: tech stack, team conventions, architecture patterns, and how the codebase evolves over time."
        },
        "mental_models": [
          {
            "id": "project-context",
            "name": "Project Context",
            "source_query": "What is the project's tech stack, architecture, and key conventions? What are the main components and how do they fit together?",
            "max_tokens": 2048,
            "trigger": {
              "refresh_after_consolidation": true
            }
          },
          {
            "id": "developer-preferences",
            "name": "Developer Preferences",
            "source_query": "What are the developer's preferences for tools, libraries, coding style, and workflow? How do they like code to be written and reviewed?",
            "max_tokens": 1024,
            "trigger": {
              "refresh_after_consolidation": true
            }
          }
        ]
      }
    },
    {
      "id": "personal-assistant",
      "name": "Personal Assistant",
      "description": "For always-on personal assistants that manage tasks, remember preferences, and maintain context across daily life. Tracks commitments, routines, and personal context.",
      "category": "assistant",
      "integrations": [
        "openclaw",
        "hermes",
        "nemoclaw",
        "hindclaw"
      ],
      "manifest": {
        "version": "1",
        "bank": {
          "retain_mission": "Extract the user's preferences, routines, scheduled events, commitments, people they mention, and any personal context they share. Track what they ask for repeatedly and what they care about.",
          "enable_observations": true,
          "observations_mission": "Track the user's stable preferences, recurring routines, important people and relationships, and how their priorities shift over time."
        },
        "mental_models": [
          {
            "id": "user-profile",
            "name": "User Profile",
            "source_query": "What do we know about this user? What are their preferences, routines, important people, and how do they like to be helped?",
            "max_tokens": 2048,
            "trigger": {
              "refresh_after_consolidation": true
            }
          },
          {
            "id": "active-tasks",
            "name": "Active Tasks & Commitments",
            "source_query": "What tasks, commitments, or follow-ups is the user currently tracking? What deadlines or promises have been made?",
            "max_tokens": 1024,
            "trigger": {
              "refresh_after_consolidation": true
            }
          }
        ]
      }
    }
  ]
}