Use OpenClaw with SaladCloud

Learn how to set up OpenClaw as a personal AI assistant connected to a self-hosted model on SaladCloud.

Last Updated: March 30, 2026

Introduction

OpenClaw is an open-source personal AI assistant with a massive ecosystem of integrations. It connects to messaging platforms including WhatsApp, Telegram, Slack, Discord, Signal, and iMessage, and supports custom model providers via OpenAI-compatible endpoints. Using OpenClaw with SaladCloud gives you a fully self-hosted AI assistant stack - your model runs on distributed GPUs, your conversations stay locally and your messaging apps if used.

OpenClaw pairs well with SaladCloud because:

• Custom provider support - configure your SaladCloud endpoint as an OpenAI-compatible provider via the config file
• Easy messaging integration - connect to WhatsApp, Telegram, Slack, Discord, and more from one assistant
• Per-hour pricing - no per-message costs for running your personal assistant

For a step-by-step guide using OpenClaw with an Ollama deployment specifically, see the OpenClaw + Ollama (Salad Hosted) + Telegram how-to guide.

Prerequisites

Before getting started, make sure you have:

• A SaladCloud account
• A messaging account for OpenClaw (a Telegram bot is the simplest starting point)

Step-by-Step Setup

Step 1: Deploy an LLM Recipe on SaladCloud

First, deploy an OpenAI-compatible LLM server on SaladCloud.

• Go to the SaladCloud portal and create an account if you do not already have one.
• Create an organization or choose an existing one, then click “Deploy a container group”.
• Select an LLM recipe. The llama.cpp Qwen3.5-35B-A3B recipe is well-suited for conversational assistant use cases. On the recipe page, provide a name and deploy - the rest is preconfigured with recommended settings.
• Once deployed, your endpoint will be live and serving an OpenAI-compatible API.

Available recipes:

Ready-to-deploy recipes (best for less technical users):

• qwen3.5-35B-A3B - A powerful Mixture of Experts model optimized for instruction-following tasks, ideal for agentic use cases.
• qwen3.5-9b-llama-cpp - Optimized for Qwen3.5 9B model

Recipes for custom deployments (best for advanced users):

• llama.cpp - Supports GGUF models
• sglang - High-performance inference
• vllm - Popular LLM serving framework
• ollama - Simple model management
• tgi - Hugging Face Text Generation Inference server

After deployment, note your API endpoint URL (e.g., https://your-endpoint.salad.cloud).

Step 2: Install OpenClaw

curl -fsSL https://openclaw.ai/install.sh | bash

Optional Docker path (if you prefer containerized local OpenClaw for additional security/isolation):

git clone https://github.com/openclaw/openclaw
cd openclaw
./docker-setup.sh

For full installation details, see the OpenClaw docs.

Step 3: Run the Onboarding Wizard

Start the onboarding flow:

openclaw onboard --install-daemon

During onboarding:

1. Accept the local-agent security warning (choose Yes only if you understand the agent can execute actions with your local user permissions).
2. Select the quick start path.
3. Skip model setup - you will configure the SaladCloud provider manually in the next step.
4. Select Telegram when OpenClaw asks you to choose channels (see Step 4 below).
5. Complete or skip the remaining optional steps.

Step 4: Connect Telegram

Telegram is the easiest channel to connect because it uses a bot token with no phone number required:

1. Open Telegram and search for @BotFather
2. Start a chat and send /newbot
3. Set a display name and a username ending in _bot (e.g., salad_assistant_bot)
4. Copy the bot token from BotFather’s confirmation message
5. Paste the token when OpenClaw prompts for it during onboarding
6. After onboarding, open your Telegram bot - it will send a pairing code. Run the command it provides:

openclaw pairing approve telegram <CODE>

If you skipped Telegram during onboarding, add it manually to ~/.openclaw/openclaw.json:

{
  channels: {
    telegram: {
      enabled: true,
      botToken: '<YOUR_BOT_TOKEN>',
      dmPolicy: 'pairing',
      groups: { '*': { requireMention: true } },
    },
  },
}

Step 5: Configure the SaladCloud Model Provider

OpenClaw does not support adding arbitrary custom providers through the onboarding wizard. Instead, configure your SaladCloud endpoint by editing ~/.openclaw/openclaw.json directly. You can use any name for the provider key - in this example we use saladcloud.

Add the following config, merging with any existing content:

{
  models: {
    providers: {
      saladcloud: {
        baseUrl: 'https://your-endpoint.salad.cloud/v1',
        apiKey: 'dummy',
        api: 'openai-completions',
        headers: {
          'Salad-Api-Key': '<YOUR_SALAD_API_KEY>',
        },
        models: [
          {
            id: 'qwen3.5-35b-a3b',
            name: 'qwen3.5-35b-a3b',
            reasoning: false,
            input: ['text'],
            cost: {
              input: 0,
              output: 0,
              cacheRead: 0,
              cacheWrite: 0,
            },
            contextWindow: 128000,
            maxTokens: 16384,
          },
        ],
      },
    },
  },
  agents: {
    defaults: {
      model: {
        primary: 'saladcloud/qwen3.5-35b-a3b',
      },
    },
  },
}

Replace https://your-endpoint.salad.cloud/v1 with your actual endpoint URL.

Use apiKey: 'dummy' - not 'ollama-local'. The string 'ollama-local' is only recognized by OpenClaw for providers with an ollama-prefixed name. For any other provider name, use 'dummy' or any non-empty string.

If your SaladCloud deployment does not require authentication, remove the headers section, but if it does - make sure to set the Salad-Api-Key header with your actual API key.

Step 6: Restart and Test

Apply the config changes:

openclaw doctor --fix
openclaw gateway restart

Then open your Telegram bot, or local tui and send a message:

“Hello! Summarize what SaladCloud is in two sentences.”

If the bot responds, your setup is complete. You can also access the local OpenClaw UI at http://127.0.0.1:18789/.

Troubleshooting

OpenClaw cannot reach the SaladCloud endpoint

• Confirm the container group is Running in the SaladCloud portal.
• Verify the baseUrl in your config ends with /v1.
• If auth is enabled, confirm the Salad-Api-Key header is set correctly.
• Test the endpoint directly with curl:

curl https://your-endpoint.salad.cloud/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -H 'Salad-Api-Key: <YOUR_SALAD_API_KEY>' \
  -d '{"model": "qwen3.5-35b-a3b", "messages": [{"role": "user", "content": "Hello"}]}'

Telegram bot is not responding

• Verify the botToken is correct in your config.
• Re-run pairing approval if needed: openclaw pairing approve telegram <CODE>
• Check logs:

openclaw logs --follow

Tips for Best Results

Use the 35B Model for Complex Tasks

For tasks that require complex instruction following and multi-step reasoning the Qwen 3.5-35B-A3B model provides significantly better results than the 9B model.

Provider Rotation

OpenClaw supports configuring multiple model providers. If you have multiple endpoints configured, you can specify which provider to use on a per-agent basis in the config file. This allows you to route different tasks to different models or endpoints as needed.

Model Recommendations

• Qwen 3.5-35B-A3B: Best for complex assistant tasks
• Qwen 3.5-9B: Suitable for simple Q&A and quick responses