What it does
An agent framework where you compose automations as directed acyclic graphs of nodes. Each node can execute shell commands, LLM calls (Claude or Codex), or named tools. A web dashboard provides visual editing, run history, and management of agents, integrations, and scheduled tasks. Built-in tools cover file I/O, HTTP requests, JSON manipulation, shell execution, and CSV processing. Integrations auto-generate tools for data sources like Postgres, SQLite, and Gmail. Agents support cron scheduling, multi-step pipelines with conditional branching, loops, and sub-agent invocation. Can be exposed as an MCP server for integration with Claude Desktop and other MCP clients.
Who it's for
Backend engineers, data engineers, and DevOps teams building complex LLM-driven automation that integrates databases, APIs, and file systems. Teams that need to orchestrate multi-step workflows with conditional logic and centralized visibility via a dashboard.
Common use cases
- Execute conditional chains of LLM calls, shell commands, and database queries
- Schedule agents to run on cron for recurring data analysis and reporting
- Integrate Postgres, SQLite, APIs, and CSV files; let Claude decide next steps
- Expose agent workflows as MCP tools for Claude Desktop and other MCP clients
- Auto-generate agent scaffolding from plain-language goals via the AI-powered planner
Setup pitfalls
- Filesystem access required for integrations — audit data source credentials and sandbox untrusted agents
- No CI/test automation — latest commit is recent, but builds and tests are not automated
- Multi-provider LLM setup: both Claude and Codex API keys must be configured if switching between providers
- Network access: Postgres, Gmail OAuth, and webhooks require firewall and proxy configuration