How qosil/ai_memory_open works
1. Repository role
- This is not a conventional product app; it is a memory-and-governance framework for AI-assisted work. The primary runtime is an AI assistant with filesystem access, and git is the persistence layer for all memory, decisions, and session continuity.
2. Core control plane
- The top-level files form the operating core: Master-AI-Context.md, setup-AI-Memory.md, staging.md, PROJECT_MEMORY_INIT.md, NextSteps.md, Key-Decisions.md, Sessions.md, and Risk-Registry.md.
- These files define how an AI starts, plans, executes, and resumes work across sessions.
3. Session lifecycle boundary
- The repository enforces a three-phase lifecycle through agents/open.md and agents/closure.md:
- Open: load context, read indexes, initialize live session files.
- Work: update session/decision logs incrementally.
- Close: finalize logs, update indexes, and persist changes back to git.
- This creates a clear boundary between transient session state and durable project memory.
4. Tiered memory model
- The README describes a 3-tier lazy-loading pattern:
- Tier 1: always-read context (Master-AI-Context.md, architecture file, module map, next steps).
- Tier 2: index files for selective lookup (Key-Decisions.md, Sessions.md, Risk-Registry.md, standards).
- Tier 3: on-demand detailed artifacts such as release plans, specs, and connector configs.
- Architecturally, this is a knowledge retrieval system optimized to avoid overloading the AI context window.
5. Persistent knowledge stores
- decisions-learnings/, session-summary/, contacts/, and the index files are the durable memory layer.
- contacts/ introduces a structured profiling subsystem for communication preferences, while the decision/session files preserve traceability of why choices were made and what happened in prior sessions.
6. Governance and standards
- Standards and Policy/ is the policy layer, with master standards for operating practice and communication.
- These standards act as cross-cutting constraints over all projects and sessions, shaping execution authority, communication style, and decision-making rules.
7. Domain expert agent library
- Memory_Agents/ is a large catalog of domain-specific advisor templates, each with an AGENT.md and a domain playbook.
- The structure includes areas such as Finance, Legal, Marketing, Operations, Sales, Strategy, Software Development, Infrastructure, Product Development, People & Culture, and more.
- These are not runtime services; they are specialized reasoning modules that can be deployed into project folders during initialization.
8. Project staging and project workspace
- projects/ is the destination for actual work, with projects/staging/ used for projects in planning or ideation.
- The repo’s lifecycle suggests a progression from setup → staging → project memory initialization, after which a project gets its own structured folder and context files.
- This is the main boundary between the reusable template system and each concrete project instance.
9. Validation and methodology artifacts
- Model_Validation/ contains white papers and planning documents that validate the underlying memory methodology and process design.
- These documents act as the research and rationale layer, supporting the operating model rather than serving as runtime inputs.
10. Zoho connector subsystem
- zoho-mcp-server/ is the only conventional software component: a Node.js MCP server that exposes Zoho integrations.
- It is organized around src/auth.js, src/index.js, src/setup.js, and tool modules for books, campaigns, crm, desk, projects, sign, and sprints.
- Its architecture is a connector boundary to external SaaS services, configured via Zoho API credentials and Claude desktop configuration.
11. External dependencies and integration points
- The repository depends on:
- Git for versioned persistence and collaboration.
- An AI assistant for reading/writing the memory files and executing the lifecycle.
- Zoho APIs for the MCP server.
- There is no central database or application backend; the filesystem and version control system are the system of record.
12. Overall system diagram implication
- A useful architecture diagram should show:
- AI assistant as the primary actor.
- Git-backed memory repository as the durable context store.
- Standards and indexes as governance and retrieval layers.
- Domain agents as reusable advisory modules.
- Project folders as instantiated workspaces.
- Zoho MCP server as the only external-service integration runtime.
Main entry points: open.md Open/Close · Memory_Agents Domain Agents · index.js Zoho MCP · CCDE-Operating-Standard.md Standards
Components 26
Memory Core
Durable context, session continuity, and retrieval indexes.
- Master Context · Top-level always-read context for starting and steering work.
Master-AI-Context.md - Setup Guide · Instructions for initializing the AI memory workflow.
setup-AI-Memory.md - Staging Flow · High-level staging and progression guide for work intake.
staging.md - Memory Init · Project-memory initialization procedure and scaffolding.
PROJECT_MEMORY_INIT.md - Next Steps · Active task queue and forward plan.
NextSteps.md - Key Decisions · Persistent record of major choices and rationale.
Key-Decisions.md - Sessions · Session continuity index and historical work trace.
Sessions.md - Risk Registry · Persistent list of project and operating risks.
Risk-Registry.md - Memory Stores · Persistent session summaries, decisions, and contact profiles.
session-summary - Contacts · Structured contact and communication preference profiles.
contacts
Governance
Operating standards, lifecycle rules, and control-plane docs.
- Open/Close · Session boundary for loading context, working, and persisting results.
agents/open.md - Closure Rules · Finalization and commit guidance for ending a session.
agents/closure.md - Standards · Operating standards and communication rules applied across work.
Standards and Policy/CCDE-Operating-Standard.md - Comm Standard · Communication norms that constrain how the assistant interacts.
Standards and Policy/Communication-Standard.md
Advisor Library
Reusable domain-specific reasoning modules.
- Domain Agents · Catalog of reusable domain-specific reasoning modules.
Memory_Agents - Strategy · Strategic reasoning module within the advisor library.
Memory_Agents/Strategy/strategy-advisor.md - Operations · Operations reasoning module within the advisor library.
Memory_Agents/Operations/operations-advisor.md - Software Dev · Software-development reasoning module within the advisor library.
Memory_Agents/Software-Development/software-development-advisor.md
Project Workspaces
Where concrete projects are staged and initialized.
- Projects · Root area for concrete project instances and initialized workspaces.
projects - Staging Area · Workspace for ideation and projects in planning.
projects/staging
Validation
Methodology research and white-paper artifacts.
- Validation Library · White papers and methodology documents supporting the operating model.
Model_Validation - Whitepaper Gen · Script that generates validation white papers.
Model_Validation/generate-whitepaper.js
Zoho Connector
Node.js MCP server for external SaaS integrations.
- Zoho MCP · Node.js server exposing Zoho tools via MCP.
zoho-mcp-server/src/index.js - Auth Setup · Zoho credential handling and authentication logic.
zoho-mcp-server/src/auth.js - Tool Modules · Zoho domain tool implementations for external service access.
zoho-mcp-server/src/tools
Other components
- AI Assistant · Primary runtime that reads, writes, and resumes work from repo memory.
Connections 29
- AI Assistant → Master Context: loads · Always reads core context before doing work.
- AI Assistant → Setup Guide: initializes · Uses setup instructions to establish the memory workflow.
- AI Assistant → Open/Close: opens · Starts each session through the open lifecycle boundary.
- AI Assistant → Closure Rules: closes · Finalizes work through the closure lifecycle boundary.
- Open/Close → Master Context: reads · Loads the always-read context during session open.
- Open/Close → Key Decisions: indexes · Pulls decision history as part of context loading.
- Open/Close → Sessions: indexes · Pulls session history for continuity.
- Open/Close → Risk Registry: indexes · Checks current risks during session open.
- AI Assistant → Next Steps: plans · Uses next steps as the active planning surface.
- AI Assistant → Standards: constrained by · Follows operating standards across all work.
- AI Assistant → Comm Standard: communicates by · Uses communication standards for interaction style.
- Domain Agents → Strategy: contains · Strategy advisor is one module in the library.
- Domain Agents → Operations: contains · Operations advisor is one module in the library.
- Domain Agents → Software Dev: contains · Software development advisor is one module in the library.
- AI Assistant → Domain Agents: deploys · Selects domain advisors on demand for specialized reasoning.
- AI Assistant → Projects: works in · Creates and updates project workspaces under projects/.
- Projects → Staging Area: starts in · New or planned work is staged here before project initialization.
- AI Assistant → Memory Stores: persists · Writes summaries and trace artifacts to durable memory stores.
- AI Assistant → Contacts: profiles · Records structured contact preferences and profiles.
- AI Assistant → Validation Library: studies · Uses validation materials to understand the methodology.
- Validation Library → Whitepaper Gen: generates · Whitepaper generation script supports the validation library.
- AI Assistant → Zoho MCP: invokes · May use the MCP server as the runtime integration boundary.
- Zoho MCP → Auth Setup: uses · Server relies on auth setup for Zoho access.
- Zoho MCP → Tool Modules: exposes · Tool modules implement Zoho service actions.
- Tool Modules → Zoho MCP: served by · Tool modules are loaded and served through the MCP runtime.
- Standards → Open/Close: governs · Standards shape how session open/close operates.
- Standards → AI Assistant: governs · Standards constrain assistant behavior.
- Projects → Memory Init: initialized by · Projects are created through the memory initialization procedure.
- Key Decisions → Sessions: cross-references · Decisions and sessions reference each other as durable memory.