gengineEA
Launch App
Beginner

Configure Claude, OpenAI, Ollama, LM Studio, or any OpenAI-compatible provider to power gengine's AI chat.

Provider Setup

gengine works with any AI provider that supports tool use. This guide covers setup for Anthropic Claude, OpenAI, Ollama, LM Studio, and custom OpenAI-compatible endpoints — including authentication, model selection, and health diagnostics.

How provider selection works

gengine does not bundle any AI subscription. You bring your own provider — either a cloud API key or a locally running model server. The MCP bridge forwards your chat messages and tool results to the provider and returns its responses to the editor.

Provider configuration lives in Resources/claude-panel/provider-settings.json. You can edit it directly or use the gear icon in the Chat tab, which writes to that file automatically.

Note: gengine uses the provider's tool use (function calling) capability to route operations. Providers that do not support tool use — such as older base-model endpoints — will not be able to execute editor operations, only generate text.

Anthropic (Claude)

Claude gives the best results with gengine because it was designed to use MCP natively. Claude 3.5 Sonnet and Claude 3.7 Sonnet are recommended for their balance of speed, tool-use accuracy, and cost.

Option A: Claude CLI (stdio transport)

The Claude CLI connects via stdio, which is the simplest setup. Install it and authenticate once:

# Install the Claude CLI globally
npm install -g @anthropic-ai/claude-cli

# Authenticate with your Anthropic account
claude auth login
# Follow the browser prompt to complete OAuth

# Verify authentication
claude --version

Once authenticated, add gengine to the Claude CLI's MCP server list. Edit your Claude CLI config at ~/.claude/claude_desktop_config.json (create it if it doesn't exist):

{
  "mcpServers": {
    "gengine": {
      "command": "node",
      "args": [
        "C:/path/to/YourProject/Plugins/gengine/Resources/mcp-bridge/dist/index.js"
      ],
      "env": {
        "GENGINE_PORT": "8080"
      }
    }
  }
}

In the gengine Chat tab, select Claude (CLI) from the provider dropdown. No API key is needed — the CLI uses your authenticated session.

Option B: Anthropic API key (HTTP transport)

If you prefer to use the Anthropic API directly (without the CLI), use the API key option:

{
  "provider": "anthropic",
  "apiKey": "sk-ant-api03-...",
  "model": "claude-sonnet-4-5",
  "maxTokens": 8192
}

Get your API key from console.anthropic.com > API Keys. Select Claude (API) in the provider dropdown, paste your key, and choose a model.

Model selection

  • claude-haiku-3-5 — fastest, lowest cost, best for simple operations
  • claude-sonnet-4-5 — recommended default; fast with excellent tool use
  • claude-opus-4 — highest capability, slower; best for complex multi-step tasks

OpenAI

GPT-4o and o3 both support tool use and work well with gengine. The gengine MCP bridge exposes an OpenAI-compatible HTTP endpoint, so the standard OpenAI SDK flow applies.

Setup

Get an API key from platform.openai.com > API keys. In the gengine Chat tab, select OpenAI from the provider dropdown and enter your key:

{
  "provider": "openai",
  "apiKey": "sk-proj-...",
  "model": "gpt-4o",
  "baseUrl": "https://api.openai.com/v1"
}

Model selection

  • gpt-4o-mini — fast and inexpensive; handles most gengine operations
  • gpt-4o — recommended; strong tool use and vision support
  • o3 — best reasoning; useful for complex Blueprint logic generation

Tip: GPT-4o includes vision support, so image paste (Ctrl+V) works out of the box with this model.

Ollama (local)

Ollama lets you run open-weight models locally at no per-token cost. It is the best option for offline use, privacy-sensitive projects, or experimentation without API billing.

Install and start Ollama

# Download Ollama from https://ollama.com and install it

# Pull a model that supports tool use
ollama pull mistral-small   # fast, good tool use
ollama pull llama3.2        # strong general capability
ollama pull qwen2.5-coder   # excellent for code tasks

# Verify Ollama is running
curl http://localhost:11434/api/tags

Configure gengine to use Ollama

Select Ollama from the provider dropdown in the Chat tab. The base URL defaults to http://localhost:11434. Choose your pulled model from the dropdown (gengine queries Ollama's model list automatically).

{
  "provider": "ollama",
  "baseUrl": "http://localhost:11434",
  "model": "mistral-small"
}

Tool use compatibility

Not all Ollama models support tool calling. Models that work well with gengine's MCP tool calls include: mistral-small, llama3.2, qwen2.5-coder, hermes3, and firefunction-v2. If a model does not support tool calling, gengine falls back to parsing tool calls from the model's text output, which is less reliable.

Warning: Ollama models run on your local hardware. A model like llama3.2 (8B parameters) requires at least 8 GB of VRAM. Larger models require more. Check Ollama's model page for hardware requirements before pulling.

LM Studio (local)

LM Studio is a desktop app for running local models with a GUI. It exposes an OpenAI-compatible API endpoint that gengine connects to.

Setup LM Studio

1. Download LM Studio from https://lmstudio.ai
2. Open LM Studio and search for a model in the Discover tab
   Recommended: Mistral 7B Instruct, Llama 3.2 8B Instruct,
                Qwen 2.5 Coder 7B
3. Download your chosen model
4. Go to the Local Server tab (≡ icon on the left sidebar)
5. Select your model from the dropdown
6. Click "Start Server"
   → Server starts on http://localhost:1234 by default

Configure gengine to use LM Studio

Select LM Studio from the provider dropdown, or select Custom (OpenAI-compatible) and enter the LM Studio server URL:

{
  "provider": "openai-compatible",
  "baseUrl": "http://localhost:1234/v1",
  "apiKey": "lm-studio",
  "model": "local-model"
}

LM Studio does not require a real API key — any non-empty string works. The model name "local-model" is the standard placeholder; LM Studio uses whichever model is loaded in its server tab regardless of the model name in the request.

Custom OpenAI-compatible endpoint

Any provider that implements the OpenAI chat completions API with tool calling support works with gengine. This includes services like Groq, Together AI, Fireworks, Mistral AI, and self-hosted deployments like vLLM or LocalAI.

Select Custom (OpenAI-compatible) from the provider dropdown and fill in:

{
  "provider": "openai-compatible",
  "baseUrl": "https://api.groq.com/openai/v1",
  "apiKey": "gsk_...",
  "model": "llama-3.3-70b-versatile"
}

The baseUrl must point to the root of the OpenAI-compatible API (the path that the SDK prefixes with /chat/completions). The apiKey is whatever credential format the provider requires.

Note: Tool use reliability varies across providers and models. Claude and GPT-4o have the most consistent tool calling behavior. If you encounter malformed tool calls with a custom provider, try a larger or instruction-tuned model from that provider.

Provider health diagnostics

The Diagnostics tab in the Command Center shows your provider's connection status. The AI Provider row shows one of:

  • Connected — provider is reachable and the API key / auth is valid
  • Auth error — provider is reachable but rejected credentials
  • Unreachable — network or firewall issue, or local server not running
  • No tool support — provider responded but does not support tool calling

You can also trigger a manual health check without opening the Diagnostics tab by clicking Test Connection in the Provider Settings modal (gear icon in Chat tab). A green checkmark confirms the provider is ready.

# Test that an endpoint accepts tool calls
curl https://api.openai.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "hi"}],
    "tools": [{"type": "function", "function": {
      "name": "test", "description": "test", "parameters": {}
    }}]
  }'
# If the response includes "tool_calls" or finishes without error, tool use works.

Switching between providers

You can switch providers at any time from the gear icon in the Chat tab. The switch takes effect immediately for the next message — no editor restart required.

Conversation history is preserved when you switch providers. The new provider receives the full message history in its context window, so it has the same context as the previous one.

A common pattern is to use a fast, cheap model (Ollama, Haiku, GPT-4o-mini) for quick lookups and simple operations, then switch to a more capable model (Claude Opus, GPT-4o, o3) for complex multi-step tasks like Blueprint authoring or animation state machine design.

Tip: Topic tabs make multi-provider workflows clean. Use one topic for quick editor tasks with a fast model, and a separate topic for deep reasoning tasks with a powerful model. Each topic remembers the last-used provider when you return to it.

← PreviousNext →