AI Avatar tool calling
for builders

Give your avatar tools to search information, call your APIs and trigger actions in your app during a conversation.

Built with Anam

Weather lookup

get_weather_report({
 presetId:"weather",
 arguments:{
  location:"Tokyo",
  units:"METRIC"}})

Recipe lookup

find_dinner_idea({
 presetId:"chef",
 arguments:{
  mainIngredient:
  "chicken"}})

Anam knowledge

{
 query:
 "How do tools work?"
}

Currency exchange

convert_travel_budget({
 presetId:"currency",
 arguments:{amount:750,
 fromCurrency:"USD",
 toCurrency:"EUR"}})
Micah, Anam tool-calling guide

Avatars that take real actions, not just talk.

Define tools and when to use them. The persona’s LLM decides when to call them. Use server tools to fetch data for contextual responses and client tools to modify browser state and guide users across your app.

1

Define

Write a tool schema: name, description, and JSON parameters.

2

Register

Create the tool in Lab or through the API, then attach it to your persona.

3

Run

Start a session. Handle client calls in your app, or let Anam run server tools.

Unmatched realism.
Try an avatar with live tools.

Meet Micah. Ask him about the weather in Tokyo or currency conversion, and watch him use a server tool to answer with live data.

Micah, Anam tool-calling guide
Micah, Anam tool-calling guide
import express from "express";

const app = express();
app.post("/session-token", async (_req, res) => {
  const response = await fetch("https://api.anam.ai/v1/auth/session-token", {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.ANAM_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      personaConfig: {
        name: "Weather guide",
        avatarId: process.env.ANAM_AVATAR_ID,
        voiceId: process.env.ANAM_VOICE_ID,
        llmId: "b4f89001-9638-4879-a9c3-02cc9f9f2004",
        systemPrompt: "Ask for a location, then use get_weather_report. " +
          "Set presetId to weather and put location and units inside arguments. " +
          "Use METRIC unless Fahrenheit is requested. Briefly acknowledge the lookup, " +
          "then answer from the result in two short sentences. Admit lookup failures.",
        tools: [{
          type: "server",
          subtype: "webhook",
          name: "get_weather_report",
          description: "Look up current weather for a place.",
          url: "https://anam.ai/api/tools/ai-avatar-playground/call",
          method: "POST",
          headers: { "Content-Type": "application/json" },
          parameters: {
            type: "object",
            properties: {
              presetId: { type: "string", enum: ["weather"] },
              arguments: {
                type: "object",
                properties: {
                  location: { type: "string" },
                  units: { type: "string", enum: ["METRIC", "IMPERIAL"] },
                },
                required: ["location", "units"],
              },
            },
            required: ["presetId", "arguments"],
          },
          awaitResponse: true,
        }],
      },
    }),
  });
  if (!response.ok) return res.status(502).json({ error: "Session setup failed" });
  const { sessionToken } = await response.json();
  res.json({ sessionToken });
});
app.listen(3000);

Connect your LLM and your tools

Imeld achieved 100% user adoption with Anam

Every member of Imeld’s training cohort requested continued access. Its interactive video role-play gives teams a place to practice real-world conversations.

Mark Temple, founder of Imeld
Mark Temple, founder of Imeld
Mark Temple, founder of Imeld

Built for your most important conversations.

SOC 2 Type II

Move from pilot to production with independently audited security controls, backed by Anam’s SOC 2 Type II report.

HIPAA

Bring a human face to patient education, intake and clinical training, with security and privacy controls that support HIPAA-compliant deployments.

GDPR

Build for European audiences with privacy safeguards, clear data-processing commitments and support for GDPR requirements.

Frequently asked questions

How do I add tools to my avatar?

Define a name, description and JSON parameter schema. Create the tool in Lab or through the API, attach its ID to your persona and explain when to use it in the prompt. Test it in a new session.

What is the difference between client and server tools?

Client tools run through handlers in your application, such as opening a panel. Server tools let Anam search a knowledge base or call a configured webhook. Your endpoint remains responsible for authentication, validation and any business action.

Can the avatar speak before and after a tool lookup?

Yes, with a compatible LLM. Anam-provided OpenAI and Azure models support a brief acknowledgment, tool call and continued answer in the same turn. Groq-hosted models do not support this pattern. For awaited client results, return the result from your registered handler.

Can I use a custom LLM or an external agent framework?

Tool support depends on your integration. A compatible custom LLM endpoint must support the required tool-calling format. With frameworks such as LiveKit or Pipecat, your agent typically owns the LLM and tools while Anam supplies the avatar and media.

Can tools use my knowledge base and a custom avatar?

Yes. Attach a knowledge tool to the persona and select the folders it should search. You can choose a stock or supported custom avatar independently of those tools. Micah combines Anam knowledge search with weather, recipe and currency lookups. These tools retrieve information; they do not access your accounts or make transactions.

How do I update a tool or its knowledge sources?

Update the tool’s schema, endpoint or selected folders, and adjust the persona prompt when its purpose changes. For documents, wait until processing is complete. Publish persona changes made in Lab, then start a new session and test the updated behavior.

What happens if a tool fails or times out?

Plan for missing arguments, unavailable services and incomplete results. Client tools with awaitResult enabled have a configurable timeout of one to six hundred seconds, defaulting to ten. Validate inputs, return clear results and tell the persona to acknowledge failure instead of inventing a successful action.