Skip to main content
Connect a provider when you want Embedder to use models from your own GitHub Copilot or ChatGPT subscription, or from a model server you run. Run /connect to connect it. Embedder switches to the new provider’s first model, and you can choose another one at any time. /connect (also /providers) opens Connect Provider in VS Code and in the terminal. Provider credentials stay on your machine, and your prompts go straight to the provider. For a model server you run yourself, see Local models. If your organization uses its own OpenAI, Anthropic, or Google API keys with Embedder’s hosted models, see Use organization API keys.

​
Connect GitHub Copilot

  1. Run /connect.
  2. Select Login with GitHub next to GitHub Copilot in VS Code, or select GitHub Copilot in the terminal.
  3. Copy the one-time code that Embedder shows. Embedder opens the GitHub verification page.
  4. Enter the code on that page and approve access.
  5. Wait for Connected.
The models listed are the ones your Copilot subscription offers.

​
Connect ChatGPT

  1. Run /connect.
  2. Select Connect next to ChatGPT Pro/Plus in VS Code, or select ChatGPT Pro/Plus in the terminal.
  3. Choose a sign-in method:
    • Browser opens a browser window on this machine.
    • Device code shows a code to enter on any device. Use it when the editor has no browser of its own, such as on a remote host, in a container, or over SSH.
  4. Complete sign-in and wait for Connected.
Browser sign-in returns to Embedder on local port 1455. If another program uses that port, use Device code instead.

​
Choose a model

After you connect a provider, Embedder switches the chat to that provider’s first model. To choose a different model:
  • VS Code: select the model control in the chat composer (Choose model). Models from connected providers and local endpoints are listed under Custom. You can also run /model or the Embedder: Switch Model command, which list each model by its ID with the provider ID beside it.
  • Terminal: run /model to open Select Model. A connected model shows its source after the name, such as GPT-5.5 (ChatGPT), (Copilot), or the local endpoint’s name, followed by the provider ID.
The list includes Embedder’s hosted models and the models from every provider you connected. /model appears when more than one model is available.

​
Disconnect a provider

  1. Run /connect.
  2. Select Disconnect next to the provider in VS Code, or select the connected provider in the terminal.
  3. Confirm.
Disconnecting removes the stored credentials from this machine and removes that provider’s models from the model list. You can reconnect at any time. If you were using one of its models, Embedder selects another one. Local models has no Disconnect. Turn off or remove its endpoints one at a time. See Manage endpoints. If a connected model disappears from the list, open /connect and check that the provider still shows as connected. Reconnect it if needed.

​
Use only your own models

Turn off Embedder’s hosted models when prompts must go only to the providers and local endpoints you connected.
  • In VS Code, run /connect and turn off Use Embedder’s hosted models.
  • In the terminal, run /connect, select Embedder’s own models, and press Space or Enter to switch it off.
The switch applies to every project on this machine. With it off:
  • The model list shows only your connected and local models.
  • Subagents and background work, such as conversation compression, use the model you selected.
  • If no provider or local endpoint responds, you have no model to chat with until you connect one.
With the switch on, a chat that uses a GitHub Copilot or ChatGPT model can still send some subagent and background work to Embedder’s hosted models. A chat that uses a local model keeps that work on the local model either way, unless you set a different model for subagents in models.json. This switch controls which models receive your prompts. It does not disconnect Embedder entirely: sign-in, telemetry, and the web and document search tools still reach Embedder. Turn the switch back on to use hosted models again.

​
Set a default model for a project

To select a particular model whenever a project opens, create .embedder/models.json in the project root:
.embedder/models.json
The value is provider/model-id: Use the provider’s exact model ID, such as gpt-5.5, not a display name such as GPT-5.5 (ChatGPT). In VS Code, /model lists models by ID. If you don’t know the exact ID, choose the model with /model and leave main unset. To choose models for subagents, add subagents with a subagent type, or default for every type you don’t list:
.embedder/models.json
inherit makes that subagent type use Embedder’s usual choice. To apply the same defaults to every project, put the file in ~/.embedder/models.json. The project file takes precedence for each setting it includes. In a project file, set main to "inherit" to ignore a global default and use ordinary model selection. Removing main from the project file falls back to the global file, if there is one. A default does not make a model available. If the pinned model is not available to you, Embedder ignores it and selects a model as usual. A pinned main model applies each time the project opens, so a model you choose with /model lasts until the next start. Reopen the project after you change the file, then check the selected model. If the expected model is not selected, check the JSON syntax, that the provider is connected, and the exact model ID. Embedder ignores a file with invalid JSON or an invalid value.

​
Use organization API keys

Your organization can have Embedder’s hosted OpenAI, Anthropic, and Google models use its own provider API keys. Embedder enables this for your organization. If you don’t see API Keys in the Embedder app, ask your Embedder administrator. Organization owners and admins manage the keys:
  1. In the Embedder app, choose API Keys in the sidebar.
  2. Under Provider Keys, select Add Key next to OpenAI, Anthropic, or Google.
  3. Paste the key, optionally enter a Label, and select Save.
  4. Select Test to check the key. Embedder reports Key is valid or Key is invalid.
To change a key, select Replace. To remove it, select the delete icon (Delete key) and confirm. After you save a key, the page shows only its last four characters. Other members can open the page but can’t add, replace, test, or delete keys. Once a key is saved, chats in your organization’s projects that use that provider’s hosted models use the key. You don’t change anything in VS Code or the terminal. These are still hosted models, so turning off Use Embedder’s hosted models also stops them.
Last modified on October 13, 2026