Frink: Open-source n8n-like automation tool for AI harnesses released

Core Functionality

Frink introduces fine-tuned automation capabilities directly within AI applications, contrasting with the loose automation (simple triggers + prompts) found in current AI harnesses. It functions as an MCP (Model Context Protocol), allowing both human users and AI agents to define and execute complex workflows.

Key Features

• Visual Flow Building: Users can create chains of nodes triggered by webhooks (e.g., Sentry alerts) or cron jobs.
• Custom Nodes: If a required node is missing, users can instruct an agent to write a JavaScript script, which becomes accessible in the UI.
• Parallel & Sequential Execution: Supports running up to 100 flows in parallel or chaining tasks via DAGs (Directed Acyclic Graphs) for sequential dependencies.
• Pre-defined Components: Includes HTTP request nodes, agent nodes, and command execution nodes.
• State Management: Flows can be paused, cancelled, queued, or limited at any time.

Availability and Roadmap

• License: MIT (Open-source).
• Current Support: Released for Linux and macOS.
• Model Support: Currently supports Claude and Codex. Support for Gemini/antigravity, grok, and other providers is listed as coming shortly.
• Upcoming Features: CLI support and Windows compatibility are listed as coming shortly.
• Repository: Available at github.com/benord-labs/frink.