> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orchestrahq.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Orchestra in Claude, ChatGPT and Gemini

> Connect Orchestra to Claude.ai, Claude Desktop, ChatGPT or Gemini with one URL and a sign-in, and use the same cited memory your coding agents and your team use.

Connect Orchestra to Claude, ChatGPT or Gemini and they read and add to the same
memory as your coding agents and your team. Ask "what did we decide about the
database?" in a chat and the answer cites what was saved, with a link. Say
"remember that the launch is on December 3" and it is saved where Claude Code,
Codex and your teammates can find it.

There is nothing to install. You add one URL and sign in to Orchestra.

```
https://api.orchestrahq.ai/mcp
```

## Personal memories and groups

For a memory of your own, or one shared with a small group outside a team
workspace, use this address instead. The steps are the same.

```
https://orchestrahq.app/mcp
```

## Claude

In Claude.ai or Claude Desktop:

1. Go to **Customize**, then **Connectors**.
2. Choose **+ Add**, then **Add custom connector**.
3. Name it Orchestra, paste the URL above, and choose **Continue**.
4. Keep the sign-in settings Claude suggests and choose **Add**.
5. Choose **Connect**. Orchestra asks you to sign in, or to create an account,
   and to pick the workspace this connection reads and writes.

The connector then works in Claude on the web, on the desktop and on your
phone.

On Claude's Free plan you can add one custom connector. On a Claude Team or
Enterprise plan, an Owner adds Orchestra first, under **Organization
settings**, then **Connectors**. Then each person finds it under
**Customize**, then **Connectors**, and chooses **Connect** to sign in with
their own Orchestra account.

## ChatGPT

ChatGPT connects to Orchestra in developer mode for now:

1. In ChatGPT on the web, turn on developer mode in **Settings**.
2. Add a new app with the URL above, and choose OAuth for its sign-in.
3. Sign in to Orchestra and pick a workspace.

ChatGPT asks you to confirm each time it saves a memory. Reading never asks.
On ChatGPT Business and Enterprise, an admin may need to turn developer mode
on for you.

## Gemini

Gemini connects to Orchestra in Developer Mode for now:

1. In Gemini on the web, go to **Settings**, then **Gemini Apps Developer
   Mode**, or open [gemini.google.com/devmode](https://gemini.google.com/devmode).
2. Choose **+ Create new**, and paste the URL above into **MCP Server URL**.
   Leave the client ID and secret empty: Gemini registers itself with
   Orchestra.
3. Sign in to Orchestra and pick a workspace.

To use Orchestra from the Gemini CLI or Antigravity CLI on your computer, run
`orchestra enable` instead (see [Orchestra in your agents](https://docs.orchestrahq.ai/mcp-server.md)).

## Bring in your Gemini history

Gemini does not save your past chats to Orchestra by itself. To bring them in
once, export them from Google Takeout and import the file:

1. At [takeout.google.com](https://takeout.google.com), choose **Deselect
   all**, then select **My Activity**.
2. Under **All activity data included**, keep only **Gemini Apps**. Under
   **Multiple formats**, set **Activity records** to **JSON**.
3. Create the export. Google emails you a link to a ZIP file.
4. Run `orchestra import-chat` with the ZIP:

   ```bash theme={null}
   orchestra import-chat ~/Downloads/takeout-20261115T000000Z-001.zip
   ```

Each conversation becomes a session in your memory, with secrets removed
first, the same as every other capture. Importing a newer export later adds
only what is new. Gemini keeps your chats in the export only while Gemini Apps
Activity is on, so turns from a time it was off are not there. The same
command imports a claude.ai or ChatGPT data export.

## Claude Code without the Orchestra CLI

If you have installed the Orchestra CLI, `orchestra enable` already gives your
coding agents Orchestra's tools (see [Orchestra in your agents](https://docs.orchestrahq.ai/mcp-server.md)).
Without it, add the same URL to Claude Code:

```bash theme={null}
claude mcp add --transport http orchestra https://api.orchestrahq.ai/mcp
```

Then run `/mcp` in Claude Code and sign in.

## What it can do

A chat connection has Orchestra's memory tools and nothing else:

| Tool | What it does |
| - | - |
| `status` | Says where things stand on a topic or project in one call: the latest decisions, open and recently closed issues, recent sessions and who is working on it, each cited. |
| `recall` | Searches the memory you can see and answers with citations. Each citation has a link. |
| `remember` | Saves one statement you ask it to keep, attributed to you. |
| `list_memories` | Lists what has been remembered. |
| `save_chat` | Saves the conversation you are in, when you ask it to. |
| `forget` | Removes a statement you saved, or a saved chat. |
| `get_me` | Says who you are signed in as and which workspace the chat uses. |
| `list_workspaces` | Lists your workspaces. |
| `use_workspace` | Switches the workspace this chat reads and writes. |

It cannot change your sessions, issues or anything else in Orchestra. `status`
reads issues and sessions for one topic, and only the ones you can see.

## What it saves

Orchestra saves only what you ask it to keep. Say "remember that..." or
"save this decision", and the chat saves that one statement, in your words.
It saves the conversation itself only when you ask for that (see below).

Every saved memory records who saved it, when, and that it came from Claude,
ChatGPT or Gemini. Your teammates see "saved by" with your name, the same as for
anything else you share.

## Save a chat

To keep the conversation you are in, ask: "save this chat", "keep this
conversation" or "remember this whole thing". The chat app writes what is
saved, and tells you what it kept:

* a title, and a summary of what the chat was about and where it ended;
* the decisions, open questions, next steps and facts worth recalling later;
* the full conversation too, only if you ask to keep all of it (up to 200 KB
  of text, about 50,000 words; a longer one is not saved, so ask for the
  summary instead).

Orchestra makes no model call of its own to do this. Secrets are removed
before anything is stored, the same as for every other capture.

Each decision and fact becomes a memory that recall finds and cites as the
chat it came from, for example "Chat in Claude, 2026-11-16". The chat itself
is kept with your sessions, marked with the app it came from, beside chats you
imported with `orchestra import-chat`.

A saved chat is private to you. Your teammates cannot see it or recall from
it, even in a team workspace, unless you share it.

Ask to save the same chat again later in the conversation, and the saved chat
is updated, not saved twice: new decisions and facts are added, and ones the
newer save no longer has are dropped.

To remove a saved chat, ask the chat app to forget it ("forget the chat we
saved"). Its summary, its conversation and every memory saved from it stop
appearing in recall and in your memory list. To remove one statement from it
and keep the rest, forget that statement.

## Which workspace it uses

You pick a workspace when you sign in. To use another one, ask in the chat:
"use my Acme workspace", or "list my workspaces". The switch lasts until you
switch again, and it applies to that app only: switching in ChatGPT does not
change Claude.

Before each call, Orchestra checks that you still belong to that workspace. If
you have left it, the call is refused and nothing is saved, and the chat goes
back to the workspace you signed in with.

## Orchestra Local

A workspace kept on your laptops ([Orchestra Local](https://docs.orchestrahq.ai/local-teams.md)) works with
the same connector. Orchestra's servers hold none of its memory, so each call
is passed to your own laptop over Orchestra's end-to-end encrypted relay, and
your laptop answers it from its own store, including what teammates shared with
you through git. Orchestra's servers pass the call on and keep nothing of it.

Turn it on, once per laptop:

* In the Orchestra app: Settings, then "Let ChatGPT and Claude reach this computer".
* Or in a terminal: `orchestra connect chat-apps on`.

It is off by default. Your laptop answers only you, and only while it is awake
and Orchestra is running. If you have several laptops with it on, the one you
used most recently answers. When none is on, the chat says "Your Orchestra
laptop is offline. Open it (or the Orchestra app) and ask again." Turn it off
with `orchestra connect chat-apps off` or the same switch, and `orchestra
doctor` shows whether it is on.

## Who sees what

A memory saved to a team workspace is shared with the people in it, under the
same rules as everything else in Orchestra (see
[Shared memory](https://docs.orchestrahq.ai/shared-memory.md)). A memory saved to your personal workspace is
yours alone. A saved chat, and every memory saved from it, is yours alone
in any workspace until you share it.

A citation link opens the memory on app.orchestrahq.ai. It asks you to sign
in, and shows the memory only if you can see it. Someone who opens a link
without access sees the same page whether the memory exists or not.

## Disconnect

Remove the connector in Claude's, ChatGPT's or Gemini's settings. What you saved stays
in Orchestra, where you can find it in Knowledge or forget it.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.