Skip to content
Flow Docs
Flow home Back to the main Flow site

What is Flow?

Flow is a local-first AI orchestration platform that keeps inference data on the workstation while AI advises and a deterministic engine executes. Engineers design flows on a node-based canvas, execute them through the orchestration engine, and receive AI-assisted interpretation and validation at the point of need.

The problem it solves

Section titled “The problem it solves”

Engineers in mainframe and adjacent technical environments work across many separate systems, such as CLI tools, log viewers, job schedulers, and monitoring dashboards. The operational processes that tie those systems together often exist only in human memory, undocumented scripts, and older knowledge that was never written down. This has a real cost:

  • Onboarding is slow, and junior engineers depend on senior staff for routine tasks.
  • People have to interpret each failure by hand, every time it happens.
  • The organization has almost no visibility into what teams actually do from day to day.

Flow turns those processes into flows. You design a flow once on a canvas, and it is also represented as text you can review. It runs with a full audit trail, and you can share it as a template. Even without AI, Flow is useful as a visual orchestration platform. It reduces context switching and lets teams share templates with each other.

The architecture: Reasoning-Augmented Orchestration

Section titled “The architecture: Reasoning-Augmented Orchestration”

Flow’s architecture is Reasoning-Augmented Orchestration: AI-assisted interpretation embedded within deterministic execution. It has three properties.

First, zero egress holds by construction rather than by configuration. Every AI model runs locally through a pluggable inference engine, no inference data leaves the workstation, and no model has access to credentials, secrets, or execution privileges. This is an inherent property of the runtime, not a deployment option you switch on.

Second, the orchestration engine keeps execution authority. It is the sole custodian of credentials and the only component that runs privileged operations, so models recommend and the engine executes under the user’s authenticated session. See Isolation Boundaries.

Third, a standardized node contract lets new AI capabilities arrive as models that conform to the catalog node interface, so new domains such as CI/CD, QA, and infrastructure need domain models rather than platform rebuilds.

Zero egress on its own exists in specialized tools, and visual orchestration on its own is widely available. The combination of the two, with a node contract that lowers the cost of adding new domains, is what sets Flow apart. See the competitive landscape.

What a flow is

Section titled “What a flow is”

A flow is an execution graph of typed nodes connected by outcome-routed edges:

  • Action nodes drive external tools through adapters (shell, filesystem, any CLI tool).
  • Utility nodes are built-in primitives: delays, logging, session memory, downloads, fan-in merge, schedule markers.
  • AI nodes invoke a model. The model is local by default, and cloud use is an opt-in carve-out.
  • Agentic nodes generate whole flows from natural-language requests and monitor runs for failures.
  • Service nodes call external REST APIs described by catalog descriptors.

Edges route on pass, fail, or always, and you can optionally guard them with when conditions. The whole graph round-trips through the Flow DSL, which is a textual, versionable representation you can review in a pull request.

The proof point

Section titled “The proof point”

The first use case on the Flow platform demonstrated end-to-end mainframe workflow orchestration built on Zowe CLI, and took first place in its hackathon division. The platform is not mainframe-specific: planned use cases in CI/CD and QA extend the same architecture across domains. Read the mainframe story.

Where to go next

Section titled “Where to go next”