MemoryDock: Git-based Long-Term Memory Service for AI Agents with Persistent Context

MemoryDock is an independent Git-backed long-term memory service designed specifically for AgentDock and multi-agent workflows, addressing the long-term memory dilemma caused by the stateless nature of AI Agents. At its core, it supports Markdown+YAML storage, memory CRUD operations, text search, and automatic Git synchronization, providing persistent context for AI Agents and bridging the critical gap from conversational AI to collaborative AI colleagues.

Current LLMs are inherently stateless—context is discarded after each conversation, which becomes a bottleneck for complex agent systems. Existing solutions have their own shortcomings: vector databases excel at semantic retrieval but lack structured storage and version control; traditional databases require complex schema design and have unnatural interactions; file systems are simple but lack search and synchronization mechanisms. MemoryDock emerges as a natural and reliable long-term memory solution.

MemoryDock is positioned as the memory layer in the agent technology stack (AgentDock as the tool execution layer, Git as the persistence and synchronization backend). Key designs: Separation of memory and execution (multi-agent sharing, state persistence, flexible deployment); Git as the storage backend (version control, audit tracking, collaborative synchronization, conflict resolution, offline-first); Markdown+YAML hybrid format (YAML metadata + Markdown content, balancing readability and parsability).

Supports memory CRUD (Create/Read/Update/Delete, with recoverable soft deletion); Two search modes (text keyword retrieval, path hierarchy browsing); Special features (project context packaging, Inbox append mode); Automatic Git synchronization (scheduled pull, debounced commit, automatic push).

Developed using Go language (high performance, static binary, concurrency-friendly, mature ecosystem); Provides HTTP RESTful API (with /v1/* prefix, optional Bearer Token authentication); Configurable via environment variables (listening address, port, storage directory, automatic synchronization, etc.), default values simplify startup, and configuration items meet customization needs.

Applicable scenarios include personal knowledge management (replacing Obsidian/Logseq backend), multi-agent collaboration systems (shared memory layer), long-term project assistants (accumulating project knowledge), and human-AI collaboration enhancement (bidirectional collaboration between human editing and AI reading).

Compared with MemGPT (vector database storage, no Git version control), Zep (PostgreSQL backend, focusing on conversation history), and file system + Git (requiring self-handling of search/API/synchronization), MemoryDock encapsulates common needs and provides a reliable and easy-to-use memory infrastructure.

MemoryDock provides a practical and elegant long-term memory solution by combining mature technologies (Git, Markdown, HTTP API). It is worth evaluating for agent developers, and its open-source nature facilitates customization and expansion. In the future, it is expected to integrate the MCP protocol, helping AI Agents evolve from conversational tools to collaborative colleagues.