DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to run a planner, architect and coder pipeline on a repo with OpenFlow

Wire role cards into a saved pipeline and run parallel agents against a local codebase.

Short answer: Install Bun 1.3 or newer and Git, then check Bun with bun --version; then run git clone https://github.com/SeeRay11/OpenFlow.git, then cd OpenFlow, then bun install; set the OPENFLOW_PROJECT environment variable to the path of the repo the agents should work in. These agents edit real files. All steps.

Who it is forA developer who wants several AI agents to split up a coding task, such as planning, designing and implementing a change, instead of prompting one agent at a time.

The problemA single agent prompt tends to mix planning, design and coding in one pass, and there is no visible structure to reuse or inspect.

The outcomeA saved, named pipeline of role nodes that runs agents in parallel against the target repo, with each step visible on the canvas.

What you need first

  • Bun 1.3 or newer (check with `bun --version`)
  • Git
  • A local repo the agents are allowed to edit, set through OPENFLOW_PROJECT
  • Optional: a provider API key for paid models, or use the free `-free` models with no key

Steps

  1. Install Bun 1.3 or newer and Git, then check Bun with bun --version.
  2. Run git clone https://github.com/SeeRay11/OpenFlow.git, then cd OpenFlow, then bun install.
  3. Set the OPENFLOW_PROJECT environment variable to the path of the repo the agents should work in. These agents edit real files.
  4. Run FLOW_MANAGE_SERVER=1 bun openflow.ts (PowerShell: $env:FLOW_MANAGE_SERVER=1; bun openflow.ts), then open http://localhost:5174.
  5. Drag role cards onto the canvas, wire them as planner to architect to coder, save the pipeline, and run it.

Prompt for your agent

Using the OpenFlow repo at https://github.com/SeeRay11/OpenFlow, set OPENFLOW_PROJECT to <target-repo-path>, start it with `FLOW_MANAGE_SERVER=1 bun openflow.ts`, and build a saved planner, architect and coder pipeline on the canvas at http://localhost:5174. Run it and report the final run log. Verify that each node ran as its own agent and that the run log lists a decision for each permission request.

Caveat. The agents write real files in OPENFLOW_PROJECT, which defaults to the OpenFlow repo itself, so point it at a repo you are willing to change.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for SeeRay11/OpenFlow and "Run a planner, architect and coder pipeline on a repo"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer