flow-like

Supports Mcp

Runs where you choose — not where a vendor tells you. Your laptop, a private server, an air-gapped factory network, a Raspberry Pi, or cloud you control. Your data stays where you put it. Local-first by default; deploy to your own AWS/GCP/Azure when you're ready. No vendor lock-in, no "upgrade to enterprise for basic autonomy."

Fast enough to run on a phone. The engine is written in Rust — native code, no garbage collector, no interpreter overhead. That's what makes real workflows viable on constrained devices, and millions of executions trivial on a server.

Fully typed, fully traceable. Every workflow tracks what data flows where and why. Type-safe input/output contracts catch errors before deployment, and every input, transformation, and output is recorded with complete lineage — replay any run. View the same workflow three ways: Process for the business, Data for analysts, Execution for engineers.

AI-native, and it runs on-device. LLMs, embeddings, vision object detection, speech-to-text, and text-to-speech all run locally — no API keys, no data leaving the machine — or in the cloud when you'd rather. Every model call is logged with full context (inputs, outputs, model version, reasoning trace), with approval gates and guardrails on every step.

Your data lives in the platform — not bolted on. Most tools make you wire up an external database, object store, and warehouse before you can do anything real. Flow-Like ships a native data layer: files, tables, and hybrid keyword + vector search on the canvas, with unlimited payload sizes and terabytes of data held right where your workflows run — no external S3, Postgres, or warehouse to provision, no glue code, no egress surprises. Share datasets, apps, and templates across your team from the same system.

Ship apps, not just automations. Package workflows as installable apps with built-in storage and a no-code UI builder. Run them offline, share them, or embed the whole editor white-labeled in your own product.