# Agent

Run inference queries with tools.

```bash
npm install @agent-smith/agent
```

Supported backends:

- [Llama.cpp](https://github.com/ggerganov/llama.cpp)
- Any server that supports an OpenAI-compatible API

## Initialize an agent

### Create a language model client

Create an `Lm` instance to connect to your inference backend:

```typescript
import { Lm } from "@agent-smith/agent";

const lm = new Lm({
    name: "llamacpp",
    serverUrl: "http://localhost:8080/v1",
});
```

For a remote OpenAI-compatible backend (e.g., OpenRouter):

```typescript
import { Lm } from "@agent-smith/agent";

const lm = new Lm({
    name: "openrouter",
    serverUrl: "https://openrouter.ai/api/v1",
    apiKey: process.env.OPENROUTER_API_KEY,
});
```

### Initialize an agent

The `Agent` constructor takes an `AgentParams` object containing the `lm` provider and optional callbacks:

```typescript
import { Agent } from "@agent-smith/agent";

const agent = new Agent({
    name: "my-agent",
    lm,
    onToken: (t) => process.stdout.write(t),
});
```

### Run an inference query

The `run()` method takes a prompt and an options object. Model and sampling parameters go in the `params` field:

```typescript
const result = await agent.run("List the planets of the solar system", {
    params: {
        model: "qwen35b",
        temperature: 0.6,
        top_k: 40,
        top_p: 0.95,
        min_p: 0,
        max_tokens: 4096,
    },
    //debug: true,
    //verbose: true,
    //tools: [get_current_weather]
});

console.log(result);
```

<a href="javascript:openLink('/libraries/agent/tools')">Next: tools</a>