Published: 2026-05-01

Nimbalyst: Visual Workspace for Codex and Claude Code With Kanban and Mermaid Diagrams

Chapters / key moments (click to jump — plays here on the page)

Nimbalyst is an open-source visual workspace built from the ground up for Codex and Claude Code. It adds a GUI layer — Kanban board, Mermaid diagrams, Excalidraw drawings, and configurable autonomy controls — on top of the CLI tools that normally lack them. Crucially, it works with both Codex and Claude Code simultaneously, not locked to either one.

Source video

"Nimbalyst: The Open-Source Visual Workspace for Building with Codex and Claude Code" by Developers Digest — Watch on YouTube →

Key Takeaways

Nimbalyst requires an existing Codex or Claude Code subscription — it uses your existing credentials and adds a visual project management interface on top.
Core visual features: Kanban board for task tracking, Mermaid diagram rendering, and Excalidraw integration for visual system design.
Autonomy levels map directly to native agent permissions: "Ask questions" (default), "Allow edits" (moderate), "Allow all" (equivalent to Codex yellow mode / Claude Code's --dangerously-skip-permissions).
Open new folders directly from the interface — no need to pre-create a project before starting; ideal for greenfield experiments.
Not locked to one agent: switch between Codex and Claude Code within the same workspace, using whichever is better suited to the task at hand.
Project management features (Kanban, planning, task tracking) cover the gap between raw CLI and fully IDE-integrated workflows.

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Reproducible steps

Choose the autonomy level up front
Three settings at project creation: <strong>ask every time</strong>, <strong>allow edits</strong>, or <strong>allow all</strong>. The last is equivalent to the permissive modes in each underlying harness — and naming that equivalence is the useful part, because it makes clear what you are actually turning on.
Use both subscriptions you already have
<strong>This is the distinguishing feature.</strong> If you have authenticated the Claude Code and Codex CLIs, the tool reaches both through those existing subscriptions — no new subscription — and shows <strong>two separate usage windows</strong> so you can see remaining headroom in each. You get the underlying agents and architecture of both, in one interface.
Plan in a versioned markdown document
The agent produces a plan with a goal, success criteria and proposed tech stack, which you can <strong>edit or delete sections of before approving</strong>. It then asks clarifying questions — which provider for sign-ups, where to deploy, how a component should behave — and folds the answers back into the plan. <strong>The plan is a markdown file you can version</strong>, so the agent's own tracking document is reviewable in git.
Work in a project-management surface, not a chat log
A Kanban board and a visual workspace are built into the editor — you can ask it to render the project architecture visually, and the agent adds tags dynamically as it works.

Gotchas

You still need active subscriptions to the underlying harnesses; this is an interface over them, not a replacement for them.
Creating a project from an empty directory is called out as a small but real convenience for greenfield experiments — a hint about who the tool is aimed at.
Diagram and board features are presentation over the same underlying agents; the quality ceiling is still whichever harness and model you route to.
No pricing discussion appears in the walkthrough.