One of the quietest superpowers in OpenClaw Easy is that the AI model is not baked into your bot. You can swap it per channel, per agent, mid-week, mid-day, mid-conversation. No config file edits, no restart, no reconnecting WhatsApp or Telegram. This guide walks you through exactly how to do it, then shows three real workflows people actually run.
What this lets you do
The short version: you can run Claude on WhatsApp, a local Llama 3.2 on Telegram, and Gemini's free tier on Discord — all from one desktop app, at the same time. You do not need to pick one provider and commit. You do not need to edit a YAML file to change minds. The model is a dropdown, and the dropdown lists every provider you have configured.
This matters because the right model depends on the channel. Slack code-review bots want a strong reasoning model. A high-volume customer chat on WhatsApp wants a cheap fast model. A private journaling Telegram wants a local model that never leaves your laptop. With OpenClaw Easy, those are three settings in three different windows, not three different bots on three different servers.
What you need
- OpenClaw Easy installed on your Mac or Windows machine.
- At least one provider: an Anthropic, OpenAI, Google, or OpenRouter API key — or Ollama installed locally for free, private inference.
- A channel already added (WhatsApp, Telegram, Slack, Discord, Feishu, or Line). If you do not have one yet, add it first — switching models afterwards is the easy part.
Step 1: Open OpenClaw Easy
Launch the desktop app. If you do not have it yet, grab the free signed installer from the download page. The app opens on the dashboard, which lists every channel you have set up and the current model attached to each.
Tip: The dashboard shows the model in the channel card. If the field says "default", that channel is using whatever the agent-level default is. If it shows a specific name like claude-sonnet-4.6, that channel has an explicit override.
Step 2: Add an AI provider
Open Settings from the sidebar, then click AI Providers. Click Add and pick the provider you want:
- Anthropic — paste a key from console.anthropic.com. Unlocks the Claude family (Opus 4.7, Sonnet 4.6, Haiku).
- OpenAI — paste a key from platform.openai.com. Unlocks GPT-5, GPT-5-mini, and the OpenAI Codex models.
- Google — paste a key from Google AI Studio. Includes the free Gemini Flash tier.
- OpenRouter — paste a key from openrouter.ai. One key, 300+ models, including DeepSeek, Mistral, Qwen, and many free options. See what is OpenRouter for the longer pitch.
- Ollama — install Ollama from ollama.com and OpenClaw Easy auto-discovers it at http://localhost:11434. No key needed.
Hit Save. The provider is live immediately — its models show up in every model dropdown in the app.
Step 3: Add another provider (optional but powerful)
This step is optional, but it is where the app's flexibility shows up. Most power users keep two or three providers configured at once. A common stack:
- Anthropic for quality work (code review, long reasoning).
- Google for cheap volume (Gemini Flash, free tier).
- Ollama for private channels that never touch the cloud.
Repeat Step 2 for each provider you want. Adding a second or third provider does not slow anything down — keys are stored locally in OpenClaw Easy's encrypted credentials directory, and a provider only does work when it is actually selected for a channel or agent.
Step 4: Pick the model on a specific channel
From the dashboard, click the channel you want to change — say, your WhatsApp number. Open Channel Settings, then click Model. The dropdown shows every model from every provider you have installed, grouped by provider name. Pick one and save.
The change takes effect on the next message. The channel does not disconnect, the QR code does not refresh, and your conversation history is preserved. The very next inbound message on that channel will be answered by the new model.
Tip: Channel-level model selection is the right place for "this channel always uses model X" decisions. Per-channel is the most common setting and is usually all you need.
Step 5: (Optional) Pick a different model per agent
If you run multiple agents on the same channel — for example a support agent and a sales agent on the same WhatsApp number — you can give each agent its own model. Open the agent's settings; the same model dropdown appears at the agent level. Whatever you set there overrides the channel default for that specific agent only.
The resolution order is simple: agent override beats channel setting, channel setting beats workspace default. If nothing is set anywhere, OpenClaw Easy falls back to the first configured provider's default model.
Common workflows
Three real configurations people run on this app today:
Cheap-first WhatsApp bot
Set the WhatsApp channel default to Gemini 2.5 Flash. It is fast, costs almost nothing, and handles 90% of customer questions ("what time do you open?", "do you ship to Italy?") perfectly. For the rare long question that needs reasoning, route to a long-context agent that uses Claude Sonnet 4.6. Result: pennies per day in token spend, premium quality only when it matters.
Privacy-first Telegram bot
Pull llama3.2 with ollama pull llama3.2, then set the Telegram channel model to Ollama → Llama 3.2. Nothing leaves your laptop. Useful for personal journaling bots, private brainstorming, anything you would not paste into ChatGPT. See running a local LLM on WhatsApp for the deeper setup walkthrough, or cloud vs. local AI chatbot for the tradeoffs.
Quality-first Slack bot
Set the Slack channel default to Claude Opus 4.7 for the code-review agent. The team posts a diff, Opus reviews it thoroughly. Set a second agent on the same channel — the general Q&A bot — to GPT-5. Both run on the same Slack workspace, both share the same conversation context, and each uses the model it is best at.
Cost — switching mid-conversation tradeoffs
One thing to be honest about: switching models mid-conversation is supported, but it is not free of consequence. OpenClaw Easy preserves your message history in the channel, and the new model picks up the thread from those stored messages. New turns work cleanly — the new model answers the next question as if it had read the conversation.
What you lose: subtle fidelity. Each model interprets the same history slightly differently. Claude might have built up an internal sense of tone or task that Gemini does not infer the same way from the raw transcript. If your conversation depended heavily on the previous model's reasoning chain or implicit assumptions, the new model may rebuild that differently.
In practice this is fine for almost every conversation. Customer chats, support questions, journaling, code reviews — none of these break when the model swaps. Long-running agentic tasks with stateful tool plans are the exception; for those, finish the task before switching.
Common gotchas
"I switched the model and the bot still uses the old one"
The model setting takes effect on the next inbound message, not on past ones. If you are testing by re-reading the last reply, you are seeing the old model's output. Send a new message in the channel — that one uses the new model. If it still does not, open Channel Settings, click Restart Channel, and try again. A restart forces the channel to re-resolve its model from settings.
"OpenRouter model not showing"
OpenRouter's catalog only appears after the key is verified and at least one credit is loaded. Open the OpenRouter setup guide for the credit-and-key walkthrough. If you already have credit and the dropdown is still empty, open Settings → AI Providers, hit Refresh next to OpenRouter, and the model list rebuilds.
"Ollama model missing"
Ollama only exposes models you have pulled locally. If you want qwen2.5, run:
Then in OpenClaw Easy, open Settings → AI Providers → Ollama and hit Refresh. The newly pulled model appears in the dropdown immediately. If Ollama itself is not detected, make sure it is running — on macOS, the menu bar icon should be visible.
Frequently asked questions
Can I use different AI models in different WhatsApp channels?
Yes. OpenClaw Easy treats the model as a per-channel setting. You can run Claude Sonnet on a personal WhatsApp number, Gemini Flash on a customer-support WhatsApp number, and a local Llama 3.2 on a third — all from the same desktop app, at the same time. See the best AI models for WhatsApp bots for picking the right one per use case.
Does switching models lose chat history?
No. Switching the model on a channel or agent does not delete past messages. New turns use the new model, and the recent conversation history is replayed as context so the new model has the thread to work with. Mid-conversation switches can lose subtle fidelity because each model interprets the same history slightly differently, but the messages themselves are intact.
How many AI providers can I add at once?
There is no hard limit. You can add Anthropic, OpenAI, Google, OpenRouter, Ollama, and any OpenAI-compatible custom endpoint side by side. Each provider exposes its full model list in the same dropdown, so adding more providers just gives you more dropdown options — it does not slow anything down.
Can I switch between cloud and local models?
Yes. As long as Ollama is installed and at least one cloud key is configured, the model dropdown shows both. You can start a channel on Claude for quality, switch it to a local Llama 3.2 the next day for privacy, and switch back later — the bot stays online the whole time.
Try OpenClaw Easy free
Model switching is something you appreciate the moment you need it. The cheap model is fine until a customer asks something hard. The local model is private until you need quality. The premium model is great until you see the bill. OpenClaw Easy lets you have all three configured, ready, and one dropdown away.
Download OpenClaw Easy for free, add the two providers you already have keys for, and watch the dropdown fill up. The right model for the right channel is the easiest performance win in any AI bot setup.