Skip to content

One graph from design to trace

flowMAS is a graph-native studio and observability environment for mutable multi-agent workflows. The graph authored in the browser is the graph validated and executed by the gMAS runtime, and the same identifiers and routes remain available during live monitoring and post-run inspection.

Get started Open the live demo

flowMAS shared workflow lifecycle

What flowMAS provides

  • Author an executable role graph

    Configure agents, prompts, models, tools, routes, conditions, weights, entry points, and exit points on one canvas.

  • Validate against the runtime

    Validation materializes the gMAS graph and derives its execution order. It is more than document or schema validation.

  • Inspect the realized workflow

    Follow typed lifecycle and topology events, then move from aggregate metrics to traces, agents, model calls, tools, memory activity, and provider usage.

Project boundaries

Component Responsibility
gMAS Graph engine, agents, schedulers, tools, and execution semantics
flowMAS Browser studio, persistence, validation API, execution control, and observability integration
gMAS Observability Self-hosted event ingestion, trace reconstruction, metrics, and Trace Explorer

flowMAS consumes gMAS through a pinned Git submodule. It does not maintain a fork of the runtime.

Start locally

git clone --recurse-submodules https://github.com/frontier-ai-next/flowMAS.git
cd flowMAS
./scripts/dev-up.sh

The development launcher starts the studio on port 3000, the API on 8000, and Trace Explorer on 8100. See Getting started for prerequisites, Docker instructions, and troubleshooting.

  • Architecture explains the shared graph boundary and event flow.
  • UI guide documents the authoring and run experience.
  • Observability covers telemetry, privacy, and deployment controls.
  • Evaluation provides complete tables, methodology, and limitations.
  • Deployment covers local Compose and production operation.