The AI assistant replies to your customers in its own words, using the data you filled in under My business. To use it you connect an AI provider with your own key, write how you want it to reply, and Yammbot creates an assistant rule.
1. Choose a provider
From menú › Inteligencia artificial (menu › Artificial intelligence), choose the provider whose AI will write the replies:
| Provider | What it’s like |
|---|---|
| Groq | Free tier, no card: around 1000 requests a day and 30 per minute per model. Very fast. |
| Google Gemini | Free tier, no card, with a daily request limit per model. On the free tier, Google may use your messages to improve its products. |
| Mistral AI | Free mode, no card, with usage and rate limits. |
| OpenRouter | Free models (ending in :free) cost nothing: 50 requests a day and 20 per minute. Which free models are available changes often. |
| Cerebras | Trial with $5 of credit for 30 days; asks you to add a card. Pay-as-you-go after that. |
| OpenAI | Pay-as-you-go: you need to add credit first. |
| Anthropic (Claude) | Pay-as-you-go. You need to add credit first. |
| DeepSeek | Pay-as-you-go and very cheap. You need to top up credit before using it. |
| Together | Pay-as-you-go, with open models. You need to add credit first. |
| xAI (Grok) | Pay-as-you-go. You need to add credit first. |
| Personalizado (Custom) | Any OpenAI-compatible service: you set the base URL, the key and the model yourself. |
The ones with a free tier don’t ask for a card, but they limit how many messages you can send per day and per minute. The paid ones charge by usage.
2. Get and paste your key
The key (API key) is like a password the provider gives you so Yammbot can use your account. Creating it is free and it’s stored only on your phone — never on our servers or in the cloud backup.
- Open the provider’s website.
- Create your account or sign in with the one you have.
- Create a new key and copy it. Some sites only show it once.
- Come back to Yammbot and paste it in the key field.
3. Choose the model
The model is the specific AI that writes the reply. If you’re not sure which one to pick, leave the recommended one — it’s at the top of the list. You can also type the exact name of a model that isn’t on the list.
4. Assistant instructions
In «Cómo debe responder» (How it should reply) you write the tone and limits for the assistant. They apply to all AI rules and you can use variables like %business_name%. Yammbot always adds, even if you don’t write it, that it must not invent prices or data and that it should say a person will answer if it doesn’t know something. You can restore the suggested text whenever you want.
Other assistant settings:
- Memoria de la conversación (Conversation memory) — how many previous messages of that chat it remembers (off, 3, 6 — recommended — or 10 messages), up to 24 hours, to understand questions like “and in red?”.
- Temperatura (Temperature) — how much the reply style varies: Precisa (Precise) sticks to your data and answers almost the same way every time, Equilibrada (Balanced) is a middle ground, Creativa (Creative) varies more. For customer service, use Precise or Balanced.
- Largo de la respuesta (Reply length) — Short, Normal or Long (the maximum tokens it can use).
- Prefijo (Prefix) — fixed text added before every reply, for example “Virtual assistant:“.
5. Create the assistant rule
For the AI to reply, you need an assistant rule. You can choose its scope:
- Todos (Any) — replies to any message using the My business data. Keep it last in your rule list so your other rules reply first.
- With another match type, it only replies to messages that match its triggers.
Fallback replies for the assistant rule:
- Respuesta si hay un error (Reply if there’s an error) — sent if the AI fails (no credit, no internet, provider limit). Empty sends nothing.
- Respuesta si se infringe una política (Reply if a policy is violated) — sent if the AI refuses to reply because of its content rules. Empty sends nothing.
Testing the connection
In Ajustes (Settings), the «Probar conexión» (Test connection) button sends a test message to your provider and shows you the response, to confirm the key and model work before using them with customers.
Common errors
| Message | What it means | What to do |
|---|---|---|
| The key is wrong or doesn’t have permission | The provider rejected your key | Copy it again from the provider’s website |
| That model doesn’t exist or your account can’t use it | The chosen model isn’t available for your account | Pick another one from the list |
| Your account has no credit | The provider requires credit and you don’t have any | Add credit on the provider’s website or choose one with a free tier |
| You’ve hit your provider’s limit for now | You went over the daily or per-minute request limit | Wait a few minutes or try tomorrow |
| The provider has a problem right now | An outage or incident on the provider’s side | Try again later |
| The provider took too long to respond | The request timed out | Try again |
| The provider blocked the reply under its content policies | The customer’s message or the reply violated the provider’s rules | Review the assistant’s instructions; use the fallback reply |
| The provider replied, but with no text | Empty response from the provider | Try another model |
| The service address isn’t valid | Custom provider only: the base URL is malformed | Check it matches the format the provider asks for (for example ending in /v1) |
| Missing the key, model or service address | Incomplete configuration | Fill in all three fields before saving |
| No connection | The phone has no internet | Check the phone’s connection |
Panel data
Replies given by the AI are counted separately in Results, in the breakdown at bot.yammbo.com/panel/resultados. The app only shows replies sent today.