Can Antigravity Use Ollama, Local Models, or BYOK?

Antigravity cannot currently use Ollama, LM Studio, your provider API key, or another custom endpoint as its core reasoning model. You choose from the models Antigravity offers. A local service can still be exposed as an MCP tool, but that does not turn the service into the agent's reasoning model.

The Short Answer

Google's official Plans documentation says Antigravity does not currently support bring-your-own-key or bring-your-own-endpoint. Its Models documentation describes a model selector for the offered reasoning models and says the additional models used by the product are not customizable.

That means a local Ollama URL such as http://localhost:11434 cannot replace the model selected under the Antigravity conversation prompt. It also means that choosing a third-party model offered by Antigravity is not BYOK: Antigravity owns that integration, not you.

Model Selection, BYOK, and API Keys Are Different Things

In Antigravity CLI, the documented /modelcommand changes the offered reasoning model, while /mcp opens the MCP manager and /usage shows quota. The CLI reference keeps those controls separate; none is a custom endpoint command.

Verify Ollama Locally Before Connecting Any Tool

If your goal is to run a local service that an MCP tool can call, set up and test Ollama independently first. Download the appropriate installer from the official Ollama download page. This guide intentionally does not ask you to pipe a remote installation script into a shell.

The current Ollama quickstart uses gemma4 as its example model. After installing, these commands only start the documented local menu, run that example, and list installed models:

ollama
ollama run gemma4
ollama ls

Ollama documents its local API base at http://localhost:11434/api. Use the following read-only request to confirm the service can list its local model tags:

curl http://localhost:11434/api/tags

If that check fails, fix Ollama before troubleshooting Antigravity or MCP. If the model name in the quickstart has changed, choose a current tag from Ollama's official model library rather than copying an old model list from a blog.

What MCP Can Do With a Local Ollama Service

Antigravity's MCP guide supports local stdio servers and remote MCP servers. It describes MCP as a way to add context and custom tools for the AI model. A local MCP server can therefore expose a narrow operation that calls Ollama's documented local API—for example, summarizing a supplied text block.

This is an inference from the two documented APIs: MCP can expose a local process as a tool, and Ollama exposes a local HTTP API. You still need an MCP server that implements the tool contract and calls Ollama. Antigravity itself does not treat the Ollama endpoint as its reasoning model.

Keep that tool narrow: define its inputs, cap the text sent to it, validate its output, and leave it disabled until you have tested the server directly. The MCP configuration can then point to the local server executable using the documented command and args fields. See the Antigravity MCP setup guide for the general configuration workflow.

Credentials in MCP headers, env, or OAuth settings authenticate the MCP server or the service it calls. They do not become model-provider credentials for the Antigravity agent.

Choose the Supported Path

1. You want a different offered Antigravity model: use the in-product model selector or /model. Check the live Models page instead of relying on a static plan-by-model table.
2. You want more usage: review your baseline quota and eligible credit overages on the Plans page. AI credits increase available usage; they do not add BYOK.
3. You want Antigravity to call a local service: expose a defined operation through an MCP server, then test it with the MCP manager.
4. You need a local model to be the primary coding agent: that is outside Antigravity's current support boundary. Ollama's official CLI reference documents its own supported integration launcher.

If your actual issue is an unavailable offered model rather than local inference, use the model-capacity and quota decision guide. For billing terminology, see the Antigravity credits explanation.

Keep Account Access Separate

Do not give an unofficial client your Antigravity login. Google's Antigravity FAQ says third-party software must not access Antigravity with that login and recommends a Vertex AI or AI Studio API key when using Gemini through a third-party coding agent. That is separate from connecting a documented MCP tool inside Antigravity.

Frequently Asked Questions

Can I point Antigravity directly at Ollama?

No. Antigravity does not currently support a custom endpoint as its reasoning model. You can only select an offered model.

Can I paste a Gemini, Anthropic, or OpenAI key into Antigravity?

Not to replace the reasoning model. An API credential may be valid inside an MCP tool's own configuration, but it only authenticates that tool or downstream service.

Can an MCP server make Ollama the main agent?

No. MCP adds callable tools and context. The Antigravity reasoning model remains responsible for deciding whether to call the Ollama-backed tool and how to use its result.

Is Ollama useful alongside Antigravity?

It can be useful when you have a specific local operation to expose through MCP. Test the Ollama API and the MCP server independently, and do not describe that arrangement as BYOK.

All Sources and Links

Google Antigravity documentation

• Plans — quota, overages, and the BYOK/endpoint boundary
• Models — reasoning-model selection and non-customizable product models
• MCP — local servers, remote servers, tools, and authentication
• CLI reference — model, MCP, and usage commands
• FAQ — third-party account access and supported API-key alternative

Ollama documentation

• Official Ollama downloads
• Ollama quickstart
• Ollama local API introduction
• Ollama CLI reference and integrations