Fergana-Labs/stash

Automatically create new skills based on past agent traces

What it solves

Stash provides a persistent, shared memory layer for AI agents, preventing the loss of context between different sessions. It solves the problem of duplicated work and lost information by allowing agents to access a unified knowledge base containing session transcripts, connected data sources, and a curated wiki of entities and concepts.

How it works

Stash creates an "agent-native Drive" where session transcripts, files, and data from connected sources (like GitHub, Slack, and Notion) are automatically streamed and indexed. Agents interact with this data through a CLI, an MCP server with approximately 70 tools, a REST API, and a virtual-filesystem shell.

Key components include:

  • Memory Curator: A scheduled agent that compiles new sessions and files into a linked wiki of concepts and entities.
  • Skills: Shareable folders containing documentation and files that can be installed into an agent's environment.
  • Connected Sources: Integrations with tools like Google Drive, Gmail, and Jira that make external data searchable for agents.
  • Agent Integration: Native hooks for coding agents such as Claude Code, Cursor, and Gemini CLI to record sessions and read context.

Who it’s for

It is designed for engineering teams and individuals who use AI coding agents and need a persistent "second brain" for their projects, including live documentation, research knowledge bases, and operational playbooks.

Highlights

  • Unified Data Access: Connects agents to GitHub, Drive, Gmail, Slack, Notion, Linear, Jira, and more.
  • Persistent Context: Records session transcripts automatically to avoid duplicating work across different agent runs.
  • Agent-Native Drive: A shared space where both humans and agents can read and write Markdown, HTML, and PDFs in real time.
  • MCP Server Support: Provides a wide array of tools for agents to query and manage the knowledge base.
  • Self-Hostable: Can be deployed via Docker Compose for private environments.
  • Flexible LLM Configuration: Supports various providers (Anthropic, OpenAI, OpenRouter) and local embedding models.

Related

  • Project
    TencentCloud/TencentDB-Agent-Memory

    A shared memory server and management hub that allows AI agents to persist and share experience, skills, and project knowledge across different sessions and frameworks.

  • Project
    mem9-ai/mem9

    A persistent memory layer for AI agents that enables shared context and recall across sessions, machines, and different agent runtimes.

  • Project
    redis/agent-memory-server

    A structured memory layer for AI agents that provides intelligent short-term session memory and persistent long-term context using a two-tier storage model.

  • Project
    doobidoo/mcp-memory-service

    A self-hosted persistent memory backend for AI agents that provides cross-session context via REST API and MCP, featuring a local knowledge graph and ONNX-powered embeddings.

  • Project
    beilusaiying/always-accompany

    A multi-purpose AI agent framework focused on context and attention mechanisms, providing layered memory and dynamic retrieval for long-term companionship, coding, and work.