---
title: "Zedflow, a typed agent flow engine"
description: "A local engine and studio for typed agent flows: each flow is a graph validated before it runs, with human approvals and a trace of every turn."
url: "https://mc-stud.io/en/open-source/zedflow"
lang: "en"
updated: "2026-10-12"
---

Experimental agent flows Rust graphs human approval

# Zedflow

A local engine and studio for typed agent flows: every flow is a graph validated before it runs, with controlled context, human approvals and a trace of every turn.

Install

`git clone https://github.com/ZEDIUM-Off/zedflow && cd zedflow && pnpm --dir web install --frozen-lockfile && pnpm --dir web build`

Young tools: read the repository README before installing.

**678** commits / 12 months

**Latest version**

v0.2.0

**stars**

0

**forks**

0

**contributor**

1

**open issues**

33

**Last commit**

3 October 2026

Commits per month. Last 12 months, each repository on its own scale.

GitHub figures taken on 12 October 2026.

**Source code**

[github.com/ZEDIUM-Off/zedflow](https://github.com/ZEDIUM-Off/zedflow)

**Licence**

Source available, non-commercial use (PolyForm Noncommercial 1.0.0)

**Status**

Experimental

## What it does

Zedflow is an engine and studio for agent harnesses. A flow is a strict, typed graph, validated before it runs. Before every model call, a context strategy composes exactly what the agent receives; state lives in typed stores, not in the prompt. Bridges connect flows to each other, and every turn is traced.

The web studio lets you draw flows, tune context strategies, launch runs and inspect each turn. The `zf` CLI validates, compiles and runs the same definitions.

The command above builds the studio; the daemon is then started with Cargo, as described in the README, with a demo provider that needs no account and no real model. Zedflow is in preview and licensed under PolyForm Noncommercial: commercial use needs a separate agreement.

[Source code](https://github.com/ZEDIUM-Off/zedflow)

## When to reach for it

- A single agent loop is no longer enough and you want to compose specialised flows.
- You need to reread what each agent received and did, turn by turn.
- Some steps must wait for a human approval on precise data.

## Let’s talk about what AI could change for you

A 30-minute discovery call, by video and with no commitment, to talk about your project, your constraints and the best next step.

[Book a call](https://calendar.app.google/VQTPrCgNvQSNwdA7A)

[contact@mc-stud.io](mailto:contact@mc-stud.io?subject=First%20call)