The AI agent builder. One sheet, every channel.
Give your agent instructions, knowledge and tools, then configure it for its channel. The same agent answers on WhatsApp, on voice calls and over the API.

Support agent
v4 draftv3 is live
Instructions
You are the support agent for {{business_name}}. Reply in the language the customer writes in.
Answer from the knowledge sources only. If the answer is not there, say so and offer a person.
Hand off when the customer asks for a refund above {{refund_limit}} or asks for a human.
Variables
- business_name
- Example Store
- refund_limit
- ₹5,000
Test it — send a message. Simulated example.
Knowledge
- PDFReturns policyReady
- URLShipping and delivery pageReady
- TextFestive offersProcessing
Tools
Channels
This sheet now reads: v4 draft — answers from 2 ready sources, 4 tools on (order lookups work), hands off above ₹5,000, live on WhatsApp Business, Voice calls, API.
Six parts, lettered on the sheet above.
- Prompts
- Write the instructions once: tone, scope, what to refuse and when to hand off.
- Variables
- Placeholders are filled in when the agent runs, so one prompt can serve more than one context.
- Knowledge
- Add text, URLs and PDFs. The agent searches them by meaning and answers from the passages it finds. Knowledge & retrieval
- Tools
- Switched on per agent. Tools run live during a conversation, on chat and on calls. Tools & automations
- Version history
- Work in a draft, commit it when it is right, and roll back to an earlier version if it is not.
- Channel deployment
- The same agent definition is then configured for WhatsApp, voice calls or the API.
From template to live, in the console.
Start an agent from the Lead qualification template, let AI rewrite its instructions, then switch it live.
Asked three ways, answered once.
A customer asks about an order in mixed Hindi and English. The agent looks it up with the same tool and gives the same answer, whether the question arrives as a message, as speech or as a request. WhatsApp Business, Voice & IVR.
Mera order kab aayega? Order 4417.
shopify_order_status
Aapka order kal dispatch hua hai. Thursday tak pahunch jayega.
0:06CallerMera order kab aayega? Order 4417.
shopify_order_status
0:08AgentAapka order kal dispatch hua hai. Thursday tak pahunch jayega.
[
{ "role": "user",
"content": "Mera order kab aayega? Order 4417." },
{ "role": "assistant",
"content": "Aapka order kal dispatch hua hai. Thursday tak pahunch jayega." }
]Tools the agent can pick up mid-conversation.
Each one is switched on per agent. Connected tools can take real actions, so give an agent only the ones its job needs.
Conversation
Hand off, tag and annotate the thread it is in.
- escalate_to_human
- tag_conversation
- update_conversation_status
- add_internal_note
Knowledge
Search your sources by meaning, and save what it learns.
- search_knowledge_base
- save_to_knowledge
WhatsApp messages
Reply with more than text.
- send_quick_reply_buttons
- send_list_menu
- send_cta_url_button
- send_location
- send_image
- send_template_message
Store
Look up a connected Shopify store.
- shopify_order_status
- shopify_product_lookup
- shopify_customer_lookup
- shopify_create_draft_order
Everyday
Small things a person would reach for.
- get_current_time
- web_search
- calculator
Your own
When the built-in set stops short.
Describe an endpoint of yours as a custom API tool, or connect supported apps and MCP servers. Tools & automations and Integrations.
Change it without breaking it.
Edits land in a draft. Commit the draft to make it the agent customers meet, and roll back to an earlier version when a change was a mistake. Test it against repeatable scenarios first.
- v4Draft. Added the refund limit variable.Draft
- v3Committed. Answers from knowledge only.Live
- v2Committed. Added order lookup.
- v1Committed. First instructions.
It knows when to stop.
You write the escalation rules. When one is met, the agent hands the conversation to your team with its history attached, so nobody asks the customer to start again. Handoffs arrive in the support desk.
escalate_to_human
- Reason
- Refund above the limit you set
- Attached
- Conversation history, tags and internal notes
- Next
- A teammate picks it up and replies in the same thread
Or build the agent from code.
Everything on the sheet is also an API call with the same key you use for model access.
curl https://api.callmissed.com/api/v1/bots \
-H "Authorization: Bearer cm_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"name": "Support Agent", "system_prompt": "You are Acme support..."}'Questions about AI agents
What is an AI agent on CallMissed?
An AI agent is a configurable assistant: a language model plus your instructions, knowledge sources and tools. It handles customer conversations on WhatsApp, voice or the API, and hands off to a person when your rules say so.
Can one agent work on WhatsApp and voice?
Yes. The same agent configuration serves WhatsApp Business conversations and real-time voice calls, with the same knowledge and the same escalation rules.
Which languages can an agent use?
Agents can converse in 22 Indian languages, including mixed Hindi and English in one message. Coverage depends on the model you choose and, for voice, on the speech models configured for the agent.
How is this different from the WhatsApp product page?
WhatsApp Business covers the channel: your number, templates, campaigns and the shared inbox. The agent builder is channel-agnostic: one agent definition that you then configure for WhatsApp, voice, web or the API.
Can agents call external tools?
Yes. Built-in tools let an agent escalate to a human, tag and update the conversation, search your knowledge, send rich WhatsApp messages (quick-reply buttons, list menus, link buttons, location), look up Shopify orders, products and customers, get the current time, run a web search and use a calculator. Agents can also call your own API. Tools run live during a conversation, on chat and on calls, and are opt-in per agent.
How do I test an agent before it talks to customers?
Use agent testing to run repeatable scenarios, inspect each result and score what matters, then compare agent variants. Simulated conversations and scores are evidence for review, not proof of real-world performance.
Start with a blank sheet.
Every account starts free with 1,000 API credits. Write the instructions, add a source, and talk to your agent before anyone else does.






