Use any AI model on WhatsApp: Gemini, Llama, DeepSeek and more through OpenRouter
Models from Google, Meta, DeepSeek, Mistral, Qwen, xAI and many other labs can answer WhatsApp Business customers when an automation calls OpenRouter on your own key. OpenRouter speaks OpenAI's chat API, so one key and one credit balance reach hundreds of models by id, such as google/gemini-3.8-flash, while the number stays in the WhatsApp Business app on your phone.
easyCoexistence lets the model you choose answer on your WhatsApp Business number: Claude or GPT on their own keys, or any model in OpenRouter's catalog on one OpenRouter key, with the chat history, reply caps and the pause when you reply from the phone handled for you.
Entries in OpenRouter's public models API on 11 November 2026, from labs including Google, Qwen, Mistral, DeepSeek, Meta and xAI.
Catalog entries that output text only on 11 November 2026; the rest can also return images, audio or video.
OpenRouter's base URL, usable with the OpenAI SDK as a drop-in replacement.
OpenRouter's error when the account or API key has insufficient credits; adding credits fixes it.
The batch-priced version of a model, used through OpenRouter's Batch API rather than live replies.
Anthropic, OpenAI and OpenRouter keys work in the Reply with AI step; OpenRouter is the route to every other lab.
How does a Gemini, Llama or DeepSeek model reach a WhatsApp chat?
Through OpenRouter, the third provider in the Reply with AI step next to Anthropic and OpenAI. With Provider set to OpenRouter, the step sends your Instructions, the five rules easyCoexistence adds and the recent chat to OpenRouter's chat completions endpoint with your key and the model id you chose. OpenRouter routes the request to a host that serves that model and returns the text, which goes out from your number like any other reply.
Everything around the model works the same as with Claude or GPT: the 24-hour window, the pause on a chat after you answer from the phone, the cap on AI replies per chat per hour, and the run log with tokens in and out.
Which models can you use?
Any id in OpenRouter's catalog, which held 466 entries on 11 November 2026. Ids are written lab/model, and OpenRouter also lists Claude and GPT ids, so one OpenRouter key can cover those too. The catalog changes as labs release and retire models, so check openrouter.ai/models for the current list. These were listed on 11 November 2026, with the number of entries each lab had:
How do you pick an id in the editor?
The Model field reads "Pick or type a model id". With an OpenRouter key saved, its list is OpenRouter's whole public catalog, and any id you type is accepted. That list is not filtered for chat use, so a few filters are yours to apply:
- Skip ids ending in :batch. OpenRouter defines them as the batch-priced version used through its Batch API, not live replies.
- Skip image or video generation models. A customer reply needs a model whose output is text.
- A :free variant has its own rate limits, so a busy day can fail runs; test it under real traffic before relying on it.
- OpenRouter also accepts routing suffixes on any id, such as :nitro to sort hosts by throughput or :floor to sort them by price.
- Ids that start with ~ are aliases that move to the newest model of a family; a pinned id keeps replies consistent.
What does saving an OpenRouter key check?
On the Automations page, open AI keys, click Add key next to OpenRouter, paste the key and click Test and save. easyCoexistence asks OpenRouter's key endpoint who the key belongs to, which spends no tokens; a refused key is not stored, and an accepted one shows only as Added, ending in its last four characters, encrypted and used only for your automations' AI replies.
Credit is the part to watch. When the balance runs out, OpenRouter answers 402, and easyCoexistence treats that like a refused key: every automation on the account that uses OpenRouter shows "Paused: your OpenRouter key was refused. It starts again by itself once you save a key that works." and you get one email. Add credits, then click Test and save again to restart them. Other errors, such as a 429 rate limit, a 502 when the model is down or a 503 when no host fits, fail only that run and email you.
OpenRouter or a direct Anthropic or OpenAI key?
Both work in the same Reply with AI step, and you can mix them across automations on one number: Claude on an Anthropic key in the catch-all, a Gemini model through OpenRouter in a keyword automation. The difference is reach against directness. OpenRouter adds one company between you and the model and gives you one bill for many labs; a direct key keeps the request and the bill with one provider:
| OpenRouter key | Direct Anthropic or OpenAI key | |
|---|---|---|
| Models reached | Many labs, including Claude and GPT ids | That provider's models only |
| Model list in the editor | The full public catalog, unfiltered | Models your key can use; OpenAI's list shows chat models only |
| Bill | One OpenRouter credit balance | Your account with that provider |
| Out of credit | 402 pauses OpenRouter automations | Anthropic's 402 pauses Claude ones; OpenAI's 429 fails each run |
| Who handles the request | OpenRouter and the host it routes to | The provider directly |
Can a model you host yourself answer?
Yes, through the Call my webhook step rather than as a fourth provider. Point it at your server or an n8n workflow; easyCoexistence posts the message and the chat's last 10 messages, signed with X-EC-Signature, and if your endpoint answers {"reply": "..."} within 20 seconds, that text is sent to the customer. Your code can call a model running on your own hardware, a fine-tuned model, or several models in turn.
The trade is that your endpoint owns the work the Reply with AI step does for you: the prompt, the reply length and the speed. The 24-hour window and the pause after you answer from the phone still apply to the reply you return.
What does a model through OpenRouter cost?
Three separate bills. OpenRouter charges your credits for the tokens of each reply at the rates it lists for that model; the Runs tab shows tokens in and out per AI step, so average tokens times those rates times expected replies gives a monthly estimate. Meta bills the messages it charges for to your own WhatsApp Business Account, per message, by category and country. easyCoexistence charges per connected number, never per AI reply, with no markup on tokens or messages. A Test run spends real tokens too.
How to set it up
- Create an OpenRouter key and add creditsIn your OpenRouter account, create an API key and add credits, so replies are not refused for balance.
- Save it under AI keysOn the Automations page, open AI keys, click Add key next to OpenRouter, paste the key and click Test and save.
- Start from the AI assistant exampleClick New automation, pick Blank, add a Reply with AI step from the step menu, and set Provider to OpenRouter.
- Choose the model idPick an id from the list or type it in lab/model form, such as google/gemini-3.8-flash, avoiding :batch and image models.
- Test before customers see itSwitch it Off, click Save, then Test and Run test with real questions. Switch it On and save when the replies read right.
Common mistakes
- Picking an id that ends in :batch. It is OpenRouter's Batch API version of the model, meant for queued jobs, not a reply a customer waits for.
- Typing a model name without its lab prefix, such as gemini-3.8-flash. OpenRouter ids are lab/model, so the request fails and the run log shows the error.
- Running a :free variant on a busy number. Free variants have their own rate limits, and runs over the limit fail instead of replying.
- Letting the OpenRouter balance reach zero. A 402 pauses every OpenRouter automation on the account until you add credits and save the key again.
We never charge for AI or mark up tokens. US$ 9 down to US$ 0.75 per number per month, falling as you add numbers, with the first 3 days free.
easyCoexistence is a multichannel API and MCP server that connects AI agents such as Claude and ChatGPT to WhatsApp, Instagram, TikTok, Facebook and more. On WhatsApp it uses Coexistence: the business keeps using its number in the WhatsApp Business app while the AI works through Meta's official API.
Frequently asked questions
Can I paste a Google Gemini API key directly?
No. The Reply with AI step takes Anthropic, OpenAI and OpenRouter keys. Gemini models are reached through OpenRouter with an id such as google/gemini-3.8-flash, or through your own code behind a Call my webhook step.
Can different automations use different models?
Yes. Provider and model are set per Reply with AI step and per AI agent, so a keyword automation can use a small, fast model while the catch-all uses a larger one, all on the same OpenRouter key.
Who sees the messages when I use OpenRouter?
The Instructions, the chat history the step reads and the new message go to OpenRouter and to the host serving the model, under their own terms. A voice note's recording goes to your OpenAI key, or to OpenRouter when that is the only one saved, to be written out. easyCoexistence keeps the conversations of numbers that come through it for the 30, 90 or 365 days you choose, 90 by default, as that context, and run logs 30 days.
Will replies change when a lab updates its model?
Not with a pinned id. Ids that start with ~ are aliases that follow the newest model of a family, so replies can shift without notice; use the dated or versioned id to keep behavior stable.
Keep reading
Ready to get started?
Set up WhatsApp Coexistence in minutes, not months. The app keeps working on the phone.
Start Free TrialVerified on