Model providers

Model providers give agents access to the models they need. You can save provider credentials once and reuse them when creating agents.

Supported providers

Your workspace may also have managed model access. If it does, the creation flow can offer provider options without requiring you to paste a personal API key.

Use your own subscription

Claude Code, Codex, Gemini CLI, and Antigravity CLI agents can run on the CLI’s own sign-in — for example a Claude Pro/Max plan, a ChatGPT plan, a Gemini Code Assist Standard or Enterprise license, or a Google account on Antigravity CLI — instead of an API key. Gemini CLI does not accept personal Google accounts (free, Google AI Pro or Ultra); Antigravity CLI does, and with one it also runs the Claude and GPT-OSS models your Google plan includes.

1. When creating the agent, choose Use your own subscription in the model provider section.
2. After the agent is created, open its terminal from the chat page and complete the CLI’s sign-in there.
3. Select Refresh status on the sign-in card. Once the sign-in is detected, the agent is ready to chat.

The sign-in stays inside the sandbox or computer the agent runs on; Manyfold stores no API key for the agent. On a sandbox it survives pause and resume and remains until the sandbox is deleted. Agents of the same framework sharing one sandbox also share its sign-in.

Add a provider key

1. Open Settings -> Model providers.
2. Select the provider.
3. Add a label, such as personal or team.
4. Paste the API key.
5. Leave Base URL empty unless you use a compatible custom endpoint.
6. Test the connection.
7. Save the provider.

When you create an agent, choose the saved provider from the creation flow.

Provider labels

Use labels that explain ownership or budget boundaries:

• personal
• engineering-team
• staging
• managed

Avoid labels that include secret values, customer names, or one-off task names.

Rotating keys

If a key is revoked or replaced:

1. Open the provider row in settings.
2. Paste the new key.
3. Test the connection.
4. Save the provider.
5. Retry any failed agent action.

Existing agents use the updated provider connection after it is saved.

Troubleshooting

• Test connection fails: confirm the key belongs to the selected provider and has model access.
• Agent creation asks for a provider: add a compatible provider for the selected framework.
• Model list is empty: check account billing, model permissions, and any custom base URL.
• Costs look unexpected: review Settings -> Usage and confirm the agent is using the intended provider.

See also

• Create your first agent
• Use the workspace
• FAQ