clawstash
ClawStash is an AI-optimized persistent storage solution for agents that stores text, code, configs, and files via MCP or REST API. It provides dynamic retrieval with token-efficient summaries first, full content on demand.
Ficha actualizada 23 sept 2026, 2:58 · upstream 23 sept 2026, 2:26 · release sin release detectada
Qué es clawstash
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
AI agents lose context between sessions without reliable persistence. ClawStash solves this by offering a robust, structured storage system designed specifically for AI agent workflows. Built around the concept of 'stashes' — persistent memory units that can hold any kind of data including code snippets, configuration files, notes, and multi-file projects.
The core value proposition centers on token efficiency: when an agent needs to retrieve information, ClawStash returns a summarized version first (using vector search over metadata), then only fetches the full content if the user requests it. This dramatically reduces API costs while maintaining high recall through full-text search capabilities.
Key design decisions include structured key-value metadata for easy querying by tag or category, comprehensive version history tracking with configurable retention policies, and a web dashboard that provides an intuitive interface for manual browsing and management of stashes alongside the MCP toolset. The system also includes GitHub backup integration to mirror all stashes into a repository, scheduled on changes or manually triggered.
The architecture supports both server-side API calls (for programmatic agent orchestration) and client-side MCP tools (for seamless integration with agents like OpenClaw, Cursor, and others). File import from disk allows agents to drop text files directly onto the editor for instant addition as file rows without manual copy-pasting.
- AI Agents storing and retrieving context across multiple sessions
- Code assistants maintaining project state between development runs
- Knowledge management systems where structured metadata is preferred over unstructured search
- Multi-agent orchestration platforms needing shared, versioned workspace data
- Documentation repositories that require both semantic search and file-level access
- Configuration management for AI agents with persistent settings across deployments
- Token-efficient retrieval via metadata-first summarization before full content fetch
- Full-text search with ranked results enabling fast discovery of stashes by content or tags
- Structured key-value metadata and tag system for organized, queryable storage
- Version history tracking with configurable snapshot retention (default 200 snapshots)
- GitHub backup integration allowing automated mirroring of all stashes to a repository
- Web dashboard provides intuitive manual browsing alongside MCP tool automation
- MCP tools return summaries first by default, reducing unnecessary API calls and costs
- Storage capacity is limited by disk space; no built-in compression or deduplication beyond file-level storage
- Full-text search requires indexing which can be expensive for very large collections of files
- Tag-based organization may become unwieldy with thousands of tags without proper indexing optimization
- No native support for binary file types (e.g., images, executables) — only text and code are supported
- Version history retention is configurable but not unlimited; older snapshots may be lost if version limit is reached
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Deploy ClawStash using Docker Compose with a Node.js image from GitHub Container Registry (ghcr.io/fo0/clawstash). The default port mapping exposes the MCP endpoint on port 3000. For production, consider running in Kubernetes or behind a load balancer for high availability. Bind mounts are required to persist data; ensure proper ownership permissions (chown) if using Docker volumes to avoid SQLite read-only errors. For GitHub integration, configure scheduled backups via webhook events or manual triggers from the web dashboard. Always set an admin password and use scoped API tokens with appropriate permissions (read/write/mcp). The MCP tools require a separate token than the REST API endpoints — ensure your agent's MCP configuration includes both scopes.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.