AI Providers
Momo AI supports multiple AI model providers, using the OpenAI-compatible API format for requests and response parsing.
Supported Providers
Cloud Providers
- OpenAI:
api.openai.com/v1— e.g. GPT-5.4, GPT-5.4-mini, etc.; API key required. - Anthropic (Claude):
api.anthropic.com/v1— uses a Messages API adapter; e.g. claude-sonnet-4-5, claude-opus-4-1. - Google AI (Gemini):
generativelanguage.googleapis.com/v1beta/openai— Gemini 3 Flash series; supports thinking mode. - DeepSeek:
api.deepseek.com/v1—deepseek-chat/deepseek-reasoner; Reasoner always performs reasoning. - Qwen:
dashscope.aliyuncs.com/compatible-mode/v1— e.g. qwen3.5-plus, qwen3.5-flash, etc.; supportsenable_thinking. - xAI:
api.x.ai/v1— grok-4, grok-4-fast, etc. - Mistral:
api.mistral.ai/v1— mistral-large-latest, mistral-small-latest, etc. - OpenRouter:
openrouter.ai/api/v1— Multi-model aggregation; supports:onlineserver-side search. - Together AI:
api.together.xyz/v1 - Hugging Face:
router.huggingface.co/v1 - Vercel AI Gateway:
ai-gateway.vercel.sh/v1 - Novita AI:
api.novita.ai/v3/openai - Chutes:
llm.chutes.ai/v1 - BigModel (Zhipu):
open.bigmodel.cn/api/paas/v4 - Moonshot (Kimi):
api.moonshot.cn/v1— kimi-k2.5, kimi-k2, etc. - MiniMax:
api.minimaxi.chat/v1— MiniMax-Text-01, etc. - Groq:
api.groq.com/openai/v1— Fast inference; default models includeopenai/gpt-oss-120b,meta-llama/llama-4-scout-17b-16e-instruct; API key from console.groq.com/keys. - NVIDIA NIM:
integrate.api.nvidia.com/v1
Custom Provider
- Custom — Enter your own Base URL and model names for self-hosted OpenAI-compatible services, enterprise gateways, or proxies.
Local Providers
- Ollama:
localhost:11434/v1— Free local models. - LM Studio:
localhost:1234/v1— GUI for managing local models. - Hermes (Beta):
127.0.0.1:8642/v1— OpenAI-compatible API Server for a local agent; default model ishermes-agent. - OpenClaw (Beta):
ws://127.0.0.1:18789— WebSocket protocol.
Hermes (Beta) Setup
Hermes connects through its OpenAI-compatible HTTP API. Like OpenClaw, it needs gateway / API Server values in the Hermes .env before Momo can connect.
API_SERVER_ENABLED=true
API_SERVER_HOST=0.0.0.0
API_SERVER_PORT=8642
API_SERVER_KEY=<your-secret-key>
API_SERVER_CORS_ORIGINS=*
- Edit the Hermes
.envvalues above. Use your own secret forAPI_SERVER_KEY; do not use the placeholder. - For remote access or direct Chrome extension access,
API_SERVER_CORS_ORIGINS=*is the simplest working setting. For stricter deployments, replace it with trusted origins only. - Restart the Hermes Gateway. By default it listens on
http://127.0.0.1:8642. - Select Hermes (beta) in Momo, keep the default Base URL or enter your remote URL, then enable the
hermes-agentmodel. - After Connect succeeds, connection state is saved locally and restores after refreshing settings or switching theme (Base URL and API key must match the last Connect).
Recommended Hermes Workflow in Momo
- Click Connect first to verify gateway availability before sending messages, so you can avoid immediate 401/403 failures.
- When switching from another provider to Hermes, starting a new chat is recommended to keep agent context clean.
- For remote Hermes servers, confirm firewall rules and port exposure (default: 8642) before debugging app-side settings.
- If the agent reply includes HTTP images (for example
MEDIA:<url>or Markdown images), Momo downloads and displays them inline; see Image Upload — Images in AI Replies. - When editing an old user message, Hermes appends a new message and resends it to the agent instead of truncating local history; see Sidebar Chat — Edit Messages.
OpenClaw (Beta) Setup
OpenClaw is a WebSocket-based local AI gateway. The setup flow differs from other providers:
- Click Connect to connect to the OpenClaw gateway.
- After a successful connection, the button shows Connected; hover or focus to reveal Disconnect.
- Connection state is saved locally; it restores after refreshing settings or switching theme (Base URL, API key, and session key must match the last Connect).
- Sessions load automatically after connecting.
- Select a session to load its chat history in the sidebar.
- The model toggle enables automatically as
openclaw.
Click Setup Guide for full server setup and device authentication instructions.
If the gateway reply includes HTTP images (image / image_url content parts, or MEDIA:<url> in text), Momo downloads and displays them inline; see Image Upload — Images in AI Replies.
When editing an old user message, OpenClaw appends a new message and resends it to the gateway instead of truncating local history; see Sidebar Chat — Edit Messages.
Configuring a Provider
- Open the sidebar and click Settings (⚙️) in the top-right corner.
- Select AI Models or the corresponding provider section from the left navigation.
- Choose the provider you want to use and enter the API key (cloud providers typically require one; local Ollama / LM Studio usually don't).
- If you need a custom URL, fill in the API endpoint (see "Custom API Endpoint" below).
- Use Test Connection to verify the key and network connectivity before starting a conversation.
Model Management
In settings, you can fully manage your models:
- Add / Delete model entries
- Rename display names
- Enable / Disable individual models
- Custom JSON: Advanced parameters or request overrides
- Prefix prompt: Attach a fixed prompt prefix to a model
Custom API Endpoint
If you're using a proxy, enterprise gateway, or self-hosted compatible service, you can override the default Base URL in the provider settings to route requests to your custom API endpoint (must still follow OpenAI-compatible path and response format). For local providers, the default port is used, but you can change it to another machine or port.