> ## Documentation Index
> Fetch the complete documentation index at: https://praison.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Chat Options

> Per-call options for Agent.chat: attachments, reasoning, task metadata.

`AgentChatOptions` is the optional second argument to `agent.chat()` — settings that apply to one turn only.

```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph LR
    P[💬 prompt] --> O[⚙️ chat options]
    O --> A[🤖 Agent turn]
    A --> R([Response])

    classDef in fill:#8B0000,stroke:#7C90A0,color:#fff
    classDef step fill:#189AB4,stroke:#7C90A0,color:#fff

    class P in
    class O,A step
    class R step
```

## Quick Start

```typescript theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
import { Agent } from 'praisonai';

const agent = new Agent({ instructions: 'You are helpful' });
await agent.chat('Summarise this', { reasoningSteps: true });
```

***

## Options

| Option | Type | Notes |
| - | - | - |
| `attachments` | `readonly string[]` | Image paths, `http(s)://` or `data:` URIs. **Images only**; other inputs warn-and-skip. Ephemeral (never enters history). |
| `reasoningSteps` | `boolean` | Forces a single non-streaming completion; beats `agent.stream` and per-call `stream: true`. |
| `taskName` | `string` | Merged into `agent_start` / `agent_complete` hook contexts; read via `agent.getTaskContext()`. |
| `taskDescription` | `string` | As `taskName`. |
| `taskId` | `string` | As `taskName`. |
| `config` | `Record<string, unknown>` | Forwarded verbatim to a managed backend's `execute()`. |
| `errorsAsNull` | `boolean` | `chat()` only (`AgentChatCallOptions`). Resolve with `null` instead of rejecting; widens the return to `Promise<string \| null>`. |

> [!NOTE]
>
> `errorsAsNull` lives on `AgentChatCallOptions`, which extends `AgentChatOptions`. Only `chat()` honours it — `start()` and `stream()` accept `AgentChatOptions` and ignore it. Even with `errorsAsNull: true`, these still reject (Python re-raises them too):
>
> | Error | Why |
> | - | - |
> | `AbortError` / aborted signal | An aborted turn is not an empty answer |
> | `InterruptedError` | Same |
> | `ToolExecutionError` | A tool failed — the caller must handle it |
> | `BudgetExceededError` | A budget stop is not a model outcome |
>
> Swallowed errors are still reported through `Logger.error`, so a failed turn is never silent.

## Examples

```typescript theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
import { Agent } from 'praisonai';

const agent = new Agent({ instructions: 'Describe and reason' });

// attachments — an image for this turn only
await agent.chat('What is in this picture?', {
  attachments: ['./photo.png', 'https://example.com/chart.png'],
});

// reasoningSteps — one non-streaming reasoning completion
await agent.chat('Prove that sqrt(2) is irrational', { reasoningSteps: true });

// task metadata — travels on the hook contexts, readable afterwards
await agent.chat('Ingest the file', {
  taskName: 'ingest',
  taskDescription: 'load the CSV',
  taskId: 'run-42',
});
console.log(agent.getTaskContext()); // { name: 'ingest', description: 'load the CSV', id: 'run-42' }

// config — forwarded to a managed backend's execute()
await agent.chat('Run this remotely', { config: { region: 'eu-west-1' } });
```

> [!NOTE]
>
> `attachments` are **images only** (`.jpg`, `.jpeg`, `.png`, `.gif`, `.webp`). A non-image or unreadable input is reported through the warning channel and skipped — the turn is never failed. On the AI SDK backend, multimodal attachments emit a "not yet honoured" parity notice.

## Related

> [card] [Agent](https://praison.ai/docs/js/agent.md)
>
> The chat() method

> [card] [Streaming](https://praison.ai/docs/js/streaming.md)
>
> Token and event streams


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.