> ## Documentation Index
> Fetch the complete documentation index at: https://docs.base44.com/llms.txt
> Use this file to discover all available pages before exploring further.
> ## Agent Instructions
> Internal links on these pages omit the .md extension. Append .md to a docs page URL, or send an Accept: text/markdown header, to get that page as markdown.
# Setting up an AI agent for your app
> Set up an AI agent for your app, write its guidelines, choose its tools and data access, add skills and memory, and connect it to WhatsApp, Telegram, or LINE.
[AI agents](https://base44.com/ai-agent-builder) turn simple chats into actions in your Base44 app. Instead of only replying with text, they update records, trigger workflows, and connect to tools so you can run more of your process through one conversation.
Each agent understands your app's data and can search the web, call backend functions, and follow detailed guidelines you define. You can shape its tone and persona, decide which tools it can use, and choose the AI model that powers its reasoning. By default, agents use an **Automatic** model that is optimized for fast, cost-effective responses, so everything keeps working the same as before unless you decide to change it.
Memory is on by default for new agents, so every interaction feels more personal and consistent from the start. Your agent can remember important details across conversations, such as preferences, past choices, or shared knowledge, and bring that context back the next time someone reaches out.
Agents automatically figure out which tool to use based on your app and request.
Examples:
* "Mark my summer vacation booking as done" → uses the **update\_task** tool.
* "What was the most visited location in Europe in 2024?" → uses the **google\_search** tool.
* "Add 15 new tasks" → uses the **insert\_task\_records** tool.
***
## Setting up an AI agent
AI agents are available in every app, so you can start creating them right away using the AI chat.
When you chat with an AI agent, Base44 uses **integration credits**. Pricing starts at around **3 integration credits per message** with the default model and then varies based on the model you select and the length of the response.
**To set up an AI agent:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Choose a suggested agent template or click **Create from scratch** to define your own.
**Tips:**
* Use a suggested template to get started quickly with a pre-built agent role and instructions you can customize.
* If you choose **Create from scratch**, describe what you want in the AI chat and the agent is built based on your input.
* After the AI creates or edits an agent, a card for it appears in the chat. Click the card to open that agent in the editor.
***
## Customizing your AI agent
Each agent has its own configuration panel with the following sections:
* **Guidelines:** Define the agent's behavior, persona, AI model, and context files.
* **Tools:** Control what data and tools the agent can use.
* **Connectors:** Let app users connect their own external accounts to the agent via OAuth.
* **Skills:** Add reusable instruction sets that give the agent specific behaviors or expertise.
* **Memory:** Decide how the agent remembers information across conversations.
* **WhatsApp:** Connect the agent to WhatsApp so people can chat from their phone.
* **Telegram:** Connect the agent to Telegram so people can chat from anywhere.
* **LINE:** Connect the agent to LINE so people can chat from the LINE app.
### Configuring guidelines
Use the **Guidelines** tab to set how your agent behaves and which AI model it uses.
**To customize your agent's guidelines:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **Guidelines**.
5. Update the **Description** to define the agent's role and high-level behavior.
6. Update **Instructions** with detailed rules your agent should follow.
7. Under **Context files**, click **Upload files** or **Choose from library** to add reference material the agent can use to answer questions. For example, product manuals, pricing documents, policies, or FAQs. You can add up to 10 files.
8. Under **AI Model**, select the model that powers your agent. Keep **Automatic** selected for fast, cost-effective responses, or choose one of the Google Gemini, OpenAI GPT, Anthropic Claude, or GLM models for more advanced reasoning.
9. Click **Save**.
For most agents you can use the **Automatic** option. It is optimized for general-purpose tasks and interactive workflows, and starts at about **3 integration credits a message**, rising with the length of the answer.
If you need more control, you can pick a different model. What a message costs depends on the model you choose and on how long the answer is, so here is how the other models compare with **Automatic**:
| Cost compared with Automatic | Models |
| - | - |
| About the same | GPT-6 Luna |
| Several times more | Gemini 3.8 Flash, GLM 5.2, Claude Sonnet 5 |
| Many times more | Claude Opus 5.5, GPT-6.1 Sol |
| The most | GPT-6 Astra, Claude Fable 5.1 |
A long answer on a cheaper model can cost more than a short answer on a pricier one, so treat this as a guide and check **Credits Used** on a message for what it actually cost. You can switch models at any time if your needs change.
If you are not sure which model to use, leave **Automatic** selected. You can always move to a more powerful model later for advanced reasoning or analysis-heavy workflows.
### Allowing anonymous access
By default, people need to log in to your app before they can use an agent. You can let visitors chat with an agent without signing in by turning on anonymous access for that agent, so anyone can start a conversation straight away.
You can view anonymous conversations in the **Users** tab in your **Agents** dashboard, just like logged-in user conversations. [See how to view agent conversations.](#viewing-who-uses-your-agents)
When anonymous access is on, visitors can chat with the agent without signing in, on the web and on connected channels (WhatsApp and Telegram). Anonymous users can't access per-user private data, so keep this off for agents that manage personal data. Anonymous conversations also draw integration credits from your workspace pool, the same as logged-in users, so only turn this on for agents you want open to everyone. The setting is off by default and you control it per agent.
**To allow anonymous access for an agent:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **Guidelines**.
5. Turn on the **Allow anonymous access** toggle.
6. Click **Save**.
### Managing tools and data access
Use the **Tools** tab to decide what your agent can do and which data it can access.
You can ask the AI chat to add or update tools for your agent. Describe what you want the agent to do and the AI configures the right tools automatically, without you having to set them up in the **Tools** tab.
**To manage an agent's tools:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **Tools**.
5. Choose which tools the agent can use. For example:
* Entity tools to create, read, update, or delete records.
* Backend functions to send notifications, start workflows, or call external APIs.
6. Adjust any tool-specific permissions so the agent only accesses the data it needs.
7. Click **Save**.
You can add or remove tools at any time as your app evolves. For example, you might start with read-only entity tools and later allow updates once you have tested the agent.
### Adding user connectors
Use the **Connectors** tab to let your app users connect their own external accounts to the agent. When a connector is enabled, the agent can create OAuth connect links so each person can authorize access to their own account in that service directly from a conversation.
For example, in a travel advisor app, each user could connect their own Google Calendar so the agent can plan and schedule itineraries on their behalf.
[Learn more about setting up connectors for app users.](/Integrations/Connectors#setting-up-custom-oauth-for-app-users)
**To add a user connector for your agent:**
1. In the AI chat, ask the agent to add a user connector, for example: `I want my app users to connect their own LinkedIn account to this agent.`
2. When the form appears, fill in the required details. These vary by provider but typically include a connector name, Client ID, Client Secret, and the scopes your app needs.
3. Follow the on-screen instructions to authorize the connector and choose the permissions your app users are asked to grant.
4. Click **Approve**.
**Tips:**
* Before filling in the form when adding a connector, make sure you have already created an OAuth app in the provider's developer console and have your Client ID, Client Secret, and required scopes ready.
* If a user connector is already set up in your workspace, it appears in the **Connectors** tab automatically. Go to **Edit Agent**, click **Connectors**, toggle it on, and click **Save**.
* The **Connectors** tab also shows which backend functions are using each connector, so you can see exactly what the agent can do with it.
### Adding agent skills
Skills are reusable instruction sets you attach to an agent to give it specific behaviors, expertise, or policies. You can attach up to 10 skills per agent.
There are 2 types of skills:
* **App Skills:** Skills you create for a specific app. Once created, they can be reused across all agents in that app.
* **Workspace Skills:** Skills shared across your workspace. [Learn more about managing workspace skills.](https://docs.base44.com/documentation/using-your-workspaces/adding-workspace-skills)
**To add an app skill to your agent:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **Skills**.
5. Click **Add Skill** under **App Skills**.
6. Choose how to create the skill:
* Select **Start from scratch** to write your own instructions.
* Select a template such as **refund-policy**, **support-escalation**, **pricing-and-plans**, or **brand-voice** to start from a ready-made structure.
7. Enter a **Skill name**, an optional **Description**, and your **Instructions**.
8. Turn on the **Enable for current agent** toggle if you want the skill to be active right away.
9. Click **Create Skill**.
10. Click **Save**.
To use a workspace skill, scroll to **Workspace Skills** and toggle on any skill you want to enable for this agent.
### Configuring memory
Agent memory lets your agent remember important information across conversations, such as preferences, key facts, or shared knowledge. You control whether memories are shared across everyone or kept specific to each person. Your agent also manages its own memory size: when its memory reaches around 50,000 characters, it automatically condenses older entries to stay within the limit, keeping the most relevant information.
Memory is turned on by default for new agents.
**To turn on memory for your agent:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **Memory**.
5. Click the **Enable Memory** toggle to turn memory on.
6. Under **Memory Scope**, choose how this agent saves memories:
* **Global & Per User:** The agent can create shared memories for everyone and personal memories for each person.
* **Global Only:** The agent creates memories that are shared across everyone who chats with it.
* **Per User Only:** The agent creates memories that are specific to each person.
7. In **Memory Instructions**, enter guidance that explains when to save memories and what to include.
8. After your agent has a few conversations, review **Saved Memories** to see what the agent has stored. Click **Refresh** to load new memories.
9. Click **Save**.
Use memory instructions to keep your agent focused. For example:
* Save global memories for product updates, company announcements, and shared knowledge.
* Save user memories for personal preferences such as communication style, tone, and specific interests.
***
## Connecting to messaging channels
Connect your agent to outside messaging apps so people can chat with it without opening your app. Each connected channel has the same capabilities as your in-app agent, and you set up and connect each one separately.
### WhatsApp
Connect your agent to WhatsApp so people can message your app's agent directly from their phone, without even opening the app.
A WhatsApp-connected agent can do everything your in-app agent does. This includes creating, reading, updating, or deleting entities in your app, triggering backend functions such as sending notifications or starting workflows, and managing your app's data, like updating tasks and adding records. If your agent can do it inside your app, it can do it in WhatsApp too.
* This feature is available for free on all plans.
* A total of 3 in-app agents can be connected to WhatsApp across all your apps. This limit is separate from [Superagent WhatsApp connections](/superagents/customizing-your-superagent#whatsapp), which have their own pool. If you try to connect more than 3, you're prompted to disconnect one before continuing.
* Each agent is assigned its own WhatsApp number.
* WhatsApp agents use **integration credits only**. No message credits are charged. Each message sent or received uses **1 integration credit** for the WhatsApp relay, plus additional integration credits for agent actions such as LLM calls, image generation, or sending emails.
* Your app visitors don't need to install anything to use WhatsApp agents. They simply tap the WhatsApp button in your app to start a chat. People must send the first message, to help prevent spam and keep them safe. Once the conversation begins, the agent can reply just like it does inside your app.
* WhatsApp conversations support images and voice messages.
* Your agent also reads shared locations and contacts. A location comes through with its coordinates, place name and address, and a contact as a name and phone number, with nothing else from the contact card.
* Your agent can tell that someone sent a sticker, but not what the sticker shows.
* When your agent needs approval to install a skill, it asks right in your WhatsApp chat with it. Reply **Yes** to install or **No** to cancel. Other actions that need your approval are handled from the Base44 agent dashboard instead.
* If your app is public, people need to log in to your app before they can use the WhatsApp agent. This makes sure the agent can identify who it is chatting with and respond based on their data. To let people chat without logging in, turn on **Allow anonymous access** in the agent's **Guidelines** tab. Anonymous users can't access per-user private data.
**To enable WhatsApp for your agent:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **WhatsApp**.
5. Click **Send to Chat** under **Add WhatsApp to Your App** to add a WhatsApp button to your app.
6. Customize the WhatsApp welcome message.
7. Click **Save**.
### Telegram
Connect your agent to Telegram so people can message it directly from Telegram, without opening your app.
A Telegram-connected agent has the same capabilities as your in-app agent. It can create, read, update, or delete records, trigger backend functions, and manage your app's data.
* All web chat features work seamlessly in Telegram, including tools and data access.
* People can send voice notes and images for natural interaction.
* Your agent also reads shared locations, stickers and contacts. A pin or named place comes through with its coordinates and address, a sticker as the emoji it stands for, and a contact as a name and phone number, with nothing else from the contact card.
* A live location gives your agent one snapshot from the moment it was shared, not continuous updates.
* Your Telegram bot is created and connected instantly with no manual configuration needed.
* To start a conversation, people open your bot in Telegram and send `/start`.
* You can add a Telegram button to your app so people can open the bot directly from your app.
**To enable Telegram for your agent:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **Telegram**.
5. Click **Connect Telegram Bot**.
6. Follow the prompts in Telegram to create and connect your bot.
7. Click **Send to Chat** under **Add Telegram to Your App** to add a Telegram button to your app.
8. Customize the Telegram welcome message.
9. Click **Save**.
### LINE
Connect your agent to LINE so people can message it directly from the LINE app, without opening your app.
A LINE-connected agent has the same capabilities as your in-app agent. It can create, read, update, or delete records, trigger backend functions, and manage your app's data.
**Notes:**
* LINE is only available in Japan.
* LINE replies are plain text only. Avoid using markdown or HTML in your welcome message or guidelines.
**To enable LINE for your agent:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click **Edit Agent** on the relevant agent.
4. Click **LINE**.
5. Scan the QR code or copy the activation code, open LINE on your phone, add the Base44 bot, and send the activation code to connect.
6. Click **Send to Chat** under **Add LINE to Your App** to add a LINE button to your app.
7. Customize the LINE welcome message.
8. Click **Save**.
Activation codes expire after 30 minutes. If yours has expired, click **Regenerate Code** to get a new one.
***
## Chatting with your agent in a group
Your WhatsApp and Telegram agents can work in group chats, so a whole team or community can talk to the same agent in one place. Each channel connects a bit differently, and LINE groups aren't supported.
**How your agent behaves in a group:**
* By default, the agent replies to every message. You can switch it to reply only when someone mentions it or replies to one of its messages. This applies to both WhatsApp and Telegram groups.
* The agent has the same capabilities in a group as it does inside your app.
* It recognizes each member as their own user, and recognizes you as the group's owner.
* All credit usage in the group is billed to you, the app owner.
### WhatsApp groups
On WhatsApp, your agent creates the group for you and shares an invite link that others use to join. You do this by chatting with the agent, not from the WhatsApp settings.
**Limits and settings:**
* Each group can include up to 8 numbers, including the agent's own number.
* Each agent can run up to 5 groups at a time.
* To change a group's name, description, or reply mode later, message the agent from inside that WhatsApp group.
**To create a WhatsApp group:**
1. Open your agent's chat, either in your app editor or in a WhatsApp chat with the agent.
2. Ask the agent to create a WhatsApp group and give it a name. You can also set a description and choose when the agent replies.
3. The agent creates the group and sends you an invite link.
4. Share the invite link so others can join.
**Prefer to send a voice note?** Anyone in the group can send one instead of typing. Your agent transcribes it and treats it as an ordinary message, so it decides whether to reply using the same rules it applies to text. It handles images, documents, videos, locations, and contacts the same way.
### Telegram groups
On Telegram, you add the agent to a group you already belong to and make it an admin. You set this up from inside Telegram itself.
**To add your agent to a Telegram group:**
1. Open your Telegram chat with the agent's bot.
2. Choose the option to add the agent to an existing group, then select a group you already belong to. Telegram only shows groups where you can add or edit admins.
3. Add the bot to the group as an admin.
4. In the group, send `/start` to finish connecting.
Manage or disconnect Telegram groups from the **Telegram** section of your agent's settings.
***
## Viewing who uses your agents
The **Users** tab in your **Agents** dashboard shows every app user who has chatted with one of your agents. For each user you can see their role, how many agents they have used, their total conversations, and credits used. You can sort the list by agents, conversations, or credits.
You can drill down from a user into their conversations with a specific agent, and open any conversation to read the full transcript.
**To view agent users:**
1. Go to **Dashboard** in your app editor.
2. Click **Agents**.
3. Click the **Users** tab.
4. Click a user row to see which agents they have chatted with and how many conversations and credits each agent used.
5. Click **View** next to an agent to see that user's full conversation list for it.
6. Click a conversation to read the full transcript, including messages, tool calls, and credits used.
**Tips:**
* Use the search bar to find someone by name or email.
* Use **Filter by** to filter by agent or role (User, Editor, or Admin).
* Use the date picker to filter by today, the last 7 days, the last 30 days, or all time.
* Click a column header to sort by agents, conversations, or credits.
***
## FAQs
Click a question below to learn more about AI agents.
This can happen if there's an empty entity in the agent's **Tools and Capabilities** settings.
**To resolve this:**
1. Open your agent's settings.
2. Click the **Tools and Capabilities** tab.
3. Either update or remove the empty entity.
4. Click **Save**.
After updating, you should be able to send messages without errors.
Yes. Agents are available on all plans, including the Free plan.
Agents are flexible powerful tools that help you:
* Automate repetitive tasks.
* Get informed, real-time answers.
* Keep your workflow inside one app.
* Pick the right tools automatically.
Go to **Dashboard**, click **Agents**, and click the **Users** tab. Click a user to see their conversations per agent, then click **View** to open the conversation list. Click any conversation to read the full transcript.
The AI model you choose affects both how your agent behaves and how many integration credits it uses.
* **Automatic** picks a fast, general-purpose model for each message and starts at about **3 credits**, rising with the length of the answer.
* **GPT-6 Luna** costs about the same as Automatic.
* **Gemini 3.8 Flash**, **GLM 5.2**, and **Claude Sonnet 5** cost several times Automatic.
* **Claude Opus 5.5** and **GPT-6.1 Sol** cost many times Automatic.
* **GPT-6 Astra** and **Claude Fable 5.1** are built for the hardest reasoning and cost the most of any option.
If you are not sure which model to use, keep **Automatic** selected so you get a good balance of speed, quality, and cost.
You can attach up to 10 skills per agent. Skills can be a mix of app skills you create yourself and workspace skills shared across your workspace.
From **Agents** in your app's dashboard, you can see the total credits used by an agent directly below its name.
App skills are created inside a specific app and can be reused across all agents in that app. Workspace skills are shared across your entire workspace and are managed in your workspace settings. You can toggle workspace skills on or off for individual agents without editing the skill itself.
Yes. A workflow with an **In-app agent** trigger runs the moment someone starts a new conversation with your agent. Use it to get notified when a customer reaches out, add them to your contacts automatically, or send yourself a summary at the end of the day.
Ask the AI chat to set it up, for example `When someone starts a conversation with my agent, email me their name and first message`. Learn more about [creating workflows](/Building-your-app/Creating-workflows#in-app-agent).
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.