rowboat/apps/docs/docs/using-rowboat/customise/custom-llms.mdx
Tushar 893ad87268
Fix/prebuilt cards updates (#263)
* updates to the twitter prebuilt cards

* update the default fallback model from gpt-4o to gpt-4.1
2025-09-16 15:29:48 +05:30

53 lines
1.9 KiB
Text

---
title: "Custom LLMs"
description: "How to use and configure custom LLMs in Rowboat."
---
<Note> This is currently only possible in the self hosted version of Rowboat</Note>
## Using custom LLM providers
By default, Rowboat uses OpenAI LLMs (gpt-4o, gpt-4.1, etc.) for both agents and copilot, when you export your OPENAI_API_KEY.
However, you can also configure custom LLM providers (e.g. LiteLLM, OpenRouter) to use any of the hundreds of available LLMs beyond OpenAI, such as Claude, DeepSeek, Ollama LLMs and so on.
<Steps>
<Step title="Set up your LLM provider">
Configure your environment variables to point to your preferred LLM backend. Example using LiteLLM:
```bash
export PROVIDER_BASE_URL=http://host.docker.internal:4000/
export PROVIDER_API_KEY=sk-1234
```
Rowboat uses <code>gpt-4.1</code> as the default model for agents and copilot. You can override these:
```bash
export PROVIDER_DEFAULT_MODEL=claude-3-7-sonnet-latest
export PROVIDER_COPILOT_MODEL=gpt-4.1
```
**Notes:**
- Copilot is optimized for <code>gpt-4o</code>/<code>gpt-4.1</code>. We strongly recommend using these models for best results.
- You can use different models for the copilot and each agent, but all must be from the same provider (e.g., LiteLLM).
- Rowboat is provider-agnostic — any backend implementing the OpenAI messages format should work.
- OpenAI-specific tools (like <code>web_search</code>) will not function with non-OpenAI providers. Remove such tools to avoid errors.
</Step>
<Step title="Clone the repository and start Rowboat Docker">
Clone the Rowboat repo and spin it up locally:
```bash
git clone git@github.com:rowboatlabs/rowboat.git
cd rowboat
docker-compose up --build
```
</Step>
<Step title="Access the app">
Once Docker is running, navigate to:
[http://localhost:3000](http://localhost:3000)
</Step>
</Steps>