What is OpenFlow?
OpenFlow is a visual builder for multi-agent workflows from a single developer, SeeRay11. You drag role cards such as planner, architect and coder onto a canvas, wire them into a pipeline, save it and run it with parallel agents. It is a fork of OpenCode: OpenFlow's own code lives in one package, packages/flow, and the agents run on an unmodified opencode serve engine underneath.
It needs Bun 1.3+ and Git, starts with one command, and opens the canvas on localhost:5174 with the engine on port 4096. Nodes can use OpenCode's free shared model tier with no key, or any provider key added in the app. It is independent of the OpenCode team.
The repo, created in August 2026, has about 170 GitHub stars and an MIT licence; the latest release is v1.3.0 (18 October 2026).
Is OpenFlow worth it?
OpenFlow is worth knowing about as an idea more than as a dependency. It puts a node canvas in front of OpenCode so a pipeline such as planner, then architect, then coder becomes something you draw, save and rerun rather than a prompt you retype. Because all of its code sits in one package on top of an unmodified OpenCode engine, upstream improvements to OpenCode flow through, which is a sensible way to build a fork.
The defaults deserve care. Permissions are set to auto, which approves each tool request as it comes, and the agents write to the directory in OPENFLOW_PROJECT, which defaults to the OpenFlow repo itself. Switching the toolbar to ask me restores prompts, and every decision is logged either way. With no key, nodes run on OpenCode's free shared models, which the README warns return 429 and 400 errors when the tier is busy.
The operational edges show its age. The README spends much of its length on freeing ports 4096 and 5174, a database is locked error when two engines share OpenCode's data directory, and restarting the engine because it caches agent configuration at boot. Pipelines saved before a recent change need their agents regenerated once. There are about 170 stars, three open issues and one maintainer, and the project's homepage is a subreddit.
For a fixed multi-step coding pipeline, a scripted OpenCode or Claude Code subagent setup is more predictable today; for general visual agent workflows, Langflow has years more work behind it. OpenFlow is pleasant to experiment with when you want to see a multi-agent coding pipeline as a diagram, and it could become more if a second maintainer appears.
Where it earns its place
A developer exploring multi-agent coding on a side project
Compare a planner, architect, coder pipeline against a single agent
Clone, bun install, set OPENFLOW_PROJECT to a throwaway repo, switch permissions to ask me, and run the pipeline on the free -free models before adding a paid key. Zero cost until a paid model is chosen.
An engineering manager running an internal AI workshop
Show a team what a multi-agent pipeline actually does step by step
Draw the pipeline on the canvas live and run it against a sample repository; the visual layout explains roles and hand-offs better than a config file, and the run log shows each permission decision.
Before you commit
- Setup
- git clone, bun install (a large first install) and FLOW_MANAGE_SERVER=1 bun openflow.ts; the canvas opens on localhost:5174 with the engine on 4096. Needs Bun 1.3+ and Git.
- Integrations
- OpenCode's engine, agents, skills and MCP configuration; any provider key OpenCode supports, added in the titlebar or imported from the OpenCode CLI; OpenCode's free shared model tier.
- Data and privacy
- Runs locally; prompts and code go to the model provider each node uses. Permissions default to auto-approve and agents write real files in OPENFLOW_PROJECT. MIT licence; no security advisories published.
- Cost at scale
- No paid edition. Free on OpenCode's shared tier with rate limits; otherwise provider tokens multiply with the number of agent nodes in a pipeline.
- Limits
- Single maintainer; only one OpenCode engine can use the shared database at a time; agent config changes need an engine restart; older saved pipelines need agents regenerated once.
What’s good
- Visual canvas for multi-agent coding pipelines
- Clean fork: own code in one package over an unmodified OpenCode engine
- Runs with no API key on OpenCode's free model tier
- Every permission decision written to a run log
What’s not
- Permissions default to auto-approve while agents edit real files
- Single maintainer, about 170 stars, created August 2026
- Port, database-lock and restart quirks dominate the README
- Free tier is shared and often rate-limited
Best for
OpenCode users who want to experiment with drawn, reusable multi-agent coding pipelines on a sandbox repository.
OpenFlow alternatives
OpenFlow: common questions
- Is OpenFlow worth it?
- A neat idea for drawing agent pipelines on top of OpenCode, but a two-month-old, one-person fork that edits real files with auto-approved permissions is a hobby tool for now. We rate it 3.2 out of 5.
- How much does OpenFlow cost?
- Free and open source (MIT); models are free on OpenCode's shared tier or billed by your provider.
- What are the downsides of OpenFlow?
- Permissions default to auto-approve while agents edit real files; single maintainer, about 170 stars, created August 2026; port, database-lock and restart quirks dominate the README; free tier is shared and often rate-limited.
- Who is OpenFlow best for?
- OpenCode users who want to experiment with drawn, reusable multi-agent coding pipelines on a sandbox repository.
- How much does OpenFlow cost for a team or in production?
- No paid edition. Free on OpenCode's shared tier with rate limits; otherwise provider tokens multiply with the number of agent nodes in a pipeline.
- Is OpenFlow safe for company data?
- Runs locally; prompts and code go to the model provider each node uses. Permissions default to auto-approve and agents write real files in OPENFLOW_PROJECT. MIT licence; no security advisories published.
- What are the best OpenFlow alternatives?
- The closest alternatives we have reviewed are OpenCode, Langflow, Vibe Kanban and CrewAI.
Answers checked Oct 31, 2026.