Comparison

Lucid Train compared with the OpenAI Codex CLI.

Lucid Train drives the Codex CLI rather than competing with it.

Signed & notarized by Apple · opens without a Gatekeeper warning

sha256 698955a0187bc039f4c74f5d05a9f10fbb27376a45788a0a241d1326b73873c7

$curl -fsSL https://lucidtrain.com/install.sh | sh

As with OpenCode, these are not exclusive: Lucid Train can spawn `codex app-server` and speak JSON-RPC to it, using the Codex CLI login you already have. Codex is also one of the two providers that can power Design mode. So the comparison is between the Codex CLI on its own and the Codex CLI inside a harness that adds an architecture diagram, a dependency and MCP audit, a Grafana connection and a uniform approval gate.

How the pairing works

Choosing Codex spawns `codex app-server`, a long-running process speaking JSON-RPC over stdio, so a session keeps its context and can be resumed rather than replayed each turn. Design mode invokes it differently, `codex exec --json` per completion with the sandbox read-only and approvals never, so a design turn can read the project and cannot write to it.

What the harness adds around it

The architecture diagram and its conversion into an implementation specification, the dependency and MCP configuration audit, dashboards through a connected Grafana, and risk-tiered approvals with reviewable diffs. Lucid Train also registers itself as an MCP server for the session, so its diagram tools arrive pre-approved as mcp__lucid__ tools rather than prompting on first use.

The subscription question

Two routes exist. Driving the Codex CLI as the engine uses OpenAI's own client with your login and raises no interop question. Separately, the bundled Lucid engine can sign in with ChatGPT directly and run on a Codex subscription, which uses undocumented endpoints and is reverse-engineered interop subject to OpenAI's terms rather than an officially supported integration. The first route is the conservative one.

When the plain CLI is the better answer

If you live in the terminal, do not need a diagram canvas and want the fewest moving parts, the Codex CLI on its own is less to run and less to learn. The harness earns its place when the codebase is unfamiliar enough that seeing its shape first is worth more than the extra application.

Side by side

	Lucid Train	Codex CLI
Can run the other	Yes, as an engine	No
Form	Desktop app and CLI	Terminal agent
Model choice	Any provider, or a CLI engine	OpenAI models
Architecture diagrams	Yes	No
Dependency and MCP audit	Yes	No
Runs on local models	Yes	No
Drives Design mode	Yes, Codex is one of two	Not applicable
Price	CLI free and open source; desktop $3/month, 7-day trial, no card	Your OpenAI plan

FAQ

Questions

Do I need an OpenAI API key?

Not for the engine route. Lucid Train spawns the Codex CLI and the CLI authenticates itself with the login you gave it.

Can Lucid Train run on local models instead?

Yes. The bundled engine runs against local Ollama with no key and no network, which the Codex CLI cannot do.

Affiliation

Lucid Train is an independent product and is not affiliated with, endorsed by, or connected to the product described on this page. All names and trademarks belong to their respective owners. Pricing and features were checked on 22 August 2026 and change without notice; check the vendor's own site before deciding.

Lucid Train vs Gemini CLI
Stated up front: Gemini CLI is not a supported engine. Gemini models are reachable another way.
Lucid Train vs OpenHands
A platform with many surfaces, against a harness built around one idea.
Lucid Train vs draw.io
One generates the diagram from your code and solves the layout. The other is a blank canvas and a shape library.
Lucid Train vs Eraser
The closest competitor in intent, and the clearest split in approach: hosted and prompt-driven on one side, local and repository-aware on the other.

Last updated 22 August 2026