flowlite

•
3
•

Schedule, orchestrate, and manage DAG workflows, scripts, and agentic pipelines via a single lightweight binary and native MCP interface.

flowlite is an ultra-lightweight job scheduler and workflow orchestrator built in Rust that packages execution, state management, and a local web dashboard into a single zero-dependency binary backed by SQLite. Designed for standard automation scripts, cron jobs, and modern AI agent pipelines, it manages file-based YAML definitions with DAG dependencies, secret isolation, idle timeouts, and automatic retry contexts. flowlite uniquely caters to agentic workflows by passing structured outputs between steps, terminating stalled tasks, and exposing a full Model Context Protocol (stdio MCP) interface, enabling LLMs to dynamically inspect, generate, trigger, and monitor pipeline runs programmatically.

主な機能

01Declarative YAML workflows with cron scheduling, DAG execution, retries, concurrency limits, and alert triggers
02Dynamic runtime fan-out allowing running tasks or agents to submit dependent child runs
03Native Model Context Protocol (MCP) stdio server for direct agent orchestration, dynamic pipeline creation, and status monitoring
04Single-binary orchestrator using SQLite with no external database, broker, or language runtime dependencies
053 GitHub stars
06Specialized agentic support including cross-task result passing, failure-context retries, and idle timeouts

ユースケース

01Running multi-step AI agent workflows where subsequent prompts rely directly on previous step results and failure logs
02Replacing heavy orchestrators like Airflow for local or self-hosted ETL and automated data pipelines
03Enabling LLM agents to autonomously submit, monitor, and manage complex system tasks and sub-pipelines over MCP