Open Ephemeris Is Now a ChatGPT App

On 14 September 2026 OpenAI approved Open Ephemeris for the ChatGPT app directory. It is listed now.

Eleven days ago I wrote that we were not in the directory and that the only route was developer mode and a pasted server URL. That section now carries an update note. The manual path still works. You no longer need it.

Here is how to install it, and here is the one example I care about most.

Installing It

Open the listing and click Install plugin. ChatGPT will ask you to sign in to Open Ephemeris. A free account is enough; there is no card step. Then, in any chat, type @Open Ephemeris and ask for what you want.

That is the whole procedure. No URL, no developer mode, no API key to copy.

The 180° Example

While testing, I asked ChatGPT, without the app, for the rising sign of someone born 15 July 1987 at 9:01 AM in Chicago. It reported that it "worked for 14 seconds" and answered 3° Pisces. Then it built a full "Big 3" on top of that answer.

The correct Ascendant for that birth is 3°02' Virgo. That is JPL DE440, Placidus houses, the local time resolved through America/Chicago to 14:01 UTC. Same inputs, one computed answer.

Pisces is exactly opposite Virgo. ChatGPT returned the Descendant, which is the point 180° away from the one that was asked for, and it did so with a confident number attached. Everything it then said about the person's rising sign was built on the wrong horizon.

This is not a knock on the model. Language models do not have a planetary ephemeris in them, and an ephemeris, a table of where things are and when, is not the kind of thing you can reason your way to. A sign is a one-in-twelve guess. A degree inside that sign is a one-in-360 guess dressed as a measurement.

With the app installed, the same question goes to the engine. It computes the answer, it does not guess at one. The number that comes back is the number, and the wheel that renders is drawn from it, with the Ascendant marked on the left.

One honest caveat. The model still writes the prose around the result, and in one of my test runs it drew the correct wheel and then mislabeled the sign in its summary. When the wheel and the sentence disagree, the wheel is right. It is drawn from the computed longitude; the sentence is the model reading it back.

What Renders in the Chat

Once installed, four things render inline in ChatGPT rather than as text:

• the natal chart wheel
• the synastry bi-wheel
• the Human Design bodygraph
• the moon phase dial

These are the same widgets that render in Claude, served by the same server, from the same build. The previous post covers why that is true and what it cost, which was very little. Nothing in that account changed with the approval. The directory listing is a front door; the house behind it is the one that was already there.

What It Costs

Nothing to start. A free account comes with a one-time grant of calculation credits, and every question you ask through the app is a computed call against that grant. The pricing page lists what the free tier includes and what the paid tiers add if you run through it.

If You Are Not on ChatGPT

Developers and Claude users, the setup for every other client, including claude.ai connectors, Claude Desktop, Claude Code, Cursor, and any Streamable HTTP client, is on the MCP setup page. Same server, same tools, same answers.