Call Model | OpenDunes Docs

Not yet available. The Agent SDK is tracked to be built — today, use the Client SDKs against the live chat completions API. The docs below describe the intended API.

callModel() is the central primitive in the OpenDunes Agent SDK. It sends messages to the model, handles streaming, dispatches any tool calls the model makes, collects tool results, and returns the complete updated item list ready for the next turn. You call it in a loop; the SDK handles everything inside each turn.

Basic usage#

TypeScript

import { OpenDunes } from "@opendunes/agent-sdk";

const client = new OpenDunes({ apiKey: process.env.OPENDUNES_API_KEY });

const result = await client.callModel({
  model: "anthropic/claude-sonnet-5",
  messages: [{ role: "user", content: "Summarise the latest news about Algeria." }],
});

console.log(result.output);

Parameters#

Parameter	Type	Required	Description
`model`	`string`	Yes	Model slug (`provider/model`). See Models.
`messages`	`Message[]`	Yes (first turn)	New user input for this turn. Prior-turn history goes in `inputItems`.
`inputItems`	`Item[]`	Multi-turn	Item list returned by the previous `callModel()` call.
`tools`	`Tool[]`	No	Tools available to the model this turn.
`stopConditions`	`StopCondition[]`	No	Conditions that end the loop. See Stop Conditions.
`dynamicParams`	`(items: Item[]) => Params`	No	Function that returns updated parameters based on item history. See Dynamic Parameters.
`stream`	`boolean`	No	Stream the response. Default `false`.
`maxTokens`	`number`	No	Token limit for this turn.
`temperature`	`number`	No	Sampling temperature (0–2).

Return value#

callModel() returns a TurnResult:

TypeScript

interface TurnResult {
  output: string;          // Final text output for this turn
  items: Item[];           // Updated item list (pass as inputItems next turn)
  usage: TokenUsage;       // Tokens used this turn
  balance: bigint;         // Remaining balance in micro-DA
  stopReason: StopReason;  // Why the turn ended
  stoppedBy?: StopCondition; // The condition that fired, when stopReason is "stop_condition_met"
}

Multi-turn loop pattern#

TypeScript

let items: Item[] = [];

for (let turn = 0; turn < 10; turn++) {
  const result = await client.callModel({
    model: "meta-llama/llama-4-maverick",
    messages: turn === 0 ? initialMessages : undefined,
    inputItems: turn > 0 ? items : undefined,
    tools: myTools,
    stopConditions: [{ type: "no_tool_calls" }],
  });

  items = result.items;

  if (result.stopReason === "stop_condition_met") break;
}

Sections#

Working with Items — the item list model.
API Reference — full parameter and type reference.
Dynamic Parameters — adaptive per-turn params.
Next Turn Params — passing context forward.
Stop Conditions — declarative loop termination.
Streaming — stream tokens and tool calls.
Text Generation — text-focused patterns.
Message Formats — supported content types.
Tools — tool registration and dispatch.
Tool Approval & State Persistence — approval hooks and state.
Example: Weather Tool — end-to-end example.
Example: Skills Loader — dynamic tool loading.

Basic usage#

TypeScript

import { OpenDunes } from "@opendunes/agent-sdk";

const client = new OpenDunes({ apiKey: process.env.OPENDUNES_API_KEY });

const result = await client.callModel({
  model: "anthropic/claude-sonnet-5",
  messages: [{ role: "user", content: "Summarise the latest news about Algeria." }],
});

console.log(result.output);

Parameters#

Parameter	Type	Required	Description
`model`	`string`	Yes	Model slug (`provider/model`). See Models.
`messages`	`Message[]`	Yes (first turn)	New user input for this turn. Prior-turn history goes in `inputItems`.
`inputItems`	`Item[]`	Multi-turn	Item list returned by the previous `callModel()` call.
`tools`	`Tool[]`	No	Tools available to the model this turn.
`stopConditions`	`StopCondition[]`	No	Conditions that end the loop. See Stop Conditions.
`dynamicParams`	`(items: Item[]) => Params`	No	Function that returns updated parameters based on item history. See Dynamic Parameters.
`stream`	`boolean`	No	Stream the response. Default `false`.
`maxTokens`	`number`	No	Token limit for this turn.
`temperature`	`number`	No	Sampling temperature (0–2).

Return value#

callModel() returns a TurnResult:

TypeScript

interface TurnResult {
  output: string;          // Final text output for this turn
  items: Item[];           // Updated item list (pass as inputItems next turn)
  usage: TokenUsage;       // Tokens used this turn
  balance: bigint;         // Remaining balance in micro-DA
  stopReason: StopReason;  // Why the turn ended
  stoppedBy?: StopCondition; // The condition that fired, when stopReason is "stop_condition_met"
}

Multi-turn loop pattern#

TypeScript

let items: Item[] = [];

for (let turn = 0; turn < 10; turn++) {
  const result = await client.callModel({
    model: "meta-llama/llama-4-maverick",
    messages: turn === 0 ? initialMessages : undefined,
    inputItems: turn > 0 ? items : undefined,
    tools: myTools,
    stopConditions: [{ type: "no_tool_calls" }],
  });

  items = result.items;

  if (result.stopReason === "stop_condition_met") break;
}

Sections#

Working with Items — the item list model.

API Reference — full parameter and type reference.

Dynamic Parameters — adaptive per-turn params.

Next Turn Params — passing context forward.

Stop Conditions — declarative loop termination.

Streaming — stream tokens and tool calls.

Text Generation — text-focused patterns.

Message Formats — supported content types.

Tools — tool registration and dispatch.

Tool Approval & State Persistence — approval hooks and state.

Example: Weather Tool — end-to-end example.

Example: Skills Loader — dynamic tool loading.