Ash AI

Ash AI is an orchestration and deployment platform that bridges the gap between Claude Agent prototypes and production environments. Developers define agents as folders (with CLAUDE.md as the required entry point, plus optional permissions, skills, and MCP tool connections), and Ash automatically generates a complete REST API service.

Core capabilities include: session-level state persistence (SQLite/Postgres, with cross-machine pause/resume and workspace storage to S3/GCS); process-level sandbox isolation (cgroups resource limits + bubblewrap filesystem isolation); typed SSE event streams with built-in backpressure; DB-backed sandbox pool management (LRU eviction, capacity limits, cross-message reuse); and a Coordinator/Runner multi-node horizontal scaling architecture.

First-class TypeScript (@ash-ai/sdk) and Python (ash-ai-sdk) SDKs are provided, along with built-in Swagger UI documentation and API key authentication. Deployment covers Docker, AWS EC2/ECS Fargate, GCP GCE, and Kubernetes Helm charts with one-click cloud deployment scripts.

Quick Start

npm install -g @ash-ai/cli
export ANTHROPIC_API_KEY=sk-ant-...
ash start

REST API Endpoints

• POST /api/sessions — Create session
• POST /api/sessions/:id/messages — Send message (SSE)
• GET /api/sessions/:id — Query session state
• POST /api/sessions/:id/pause — Pause session
• POST /api/sessions/:id/resume — Resume session
• DELETE /api/sessions/:id — End session

Running Modes

• Standalone: Server manages sandbox child processes in-process, zero-config startup
• Coordinator: Server acts as pure control plane; Runner nodes register and host sandboxes, sessions routed by load

Unconfirmed: Team background not publicly disclosed; PyPI package status not directly verified; sandbox isolation (cgroups + bubblewrap) is Linux-specific, no stated fallback for macOS/Windows; version maturity inconsistency (repo v0.1.0 but 245 releases shown).