LLMist

Defined in: core/client.ts:59

Constructors

Constructor

new LLMist(): LLMist

Defined in: core/client.ts:71

Returns

LLMist

Constructor

new LLMist(adapters): LLMist

Defined in: core/client.ts:72

Parameters

adapters

ProviderAdapter[]

Returns

LLMist

Constructor

new LLMist(adapters, defaultProvider): LLMist

Defined in: core/client.ts:73

Parameters

adapters

ProviderAdapter[]

defaultProvider

string

Returns

LLMist

Constructor

new LLMist(options): LLMist

Defined in: core/client.ts:74

Parameters

options

LLMistOptions

Returns

LLMist

Properties

image

readonly image: ImageNamespace

Defined in: core/client.ts:67

---

modelRegistry

readonly modelRegistry: ModelRegistry

Defined in: core/client.ts:62

---

speech

readonly speech: SpeechNamespace

Defined in: core/client.ts:68

---

text

readonly text: TextNamespace

Defined in: core/client.ts:66

---

vision

readonly vision: VisionNamespace

Defined in: core/client.ts:69

Methods

complete()

complete(prompt, options?): Promise<string>

Defined in: core/client.ts:264

Instance method: Quick completion using this client instance.

Parameters

prompt

string

User prompt

options?

TextGenerationOptions

Optional configuration

Returns

Promise<string>

Complete text response

---

countTokens()

countTokens(model, messages): Promise<number>

Defined in: core/client.ts:181

Count tokens in messages for a given model.

Uses provider-specific token counting methods for accurate estimation:

• OpenAI: tiktoken library with model-specific encodings
• Anthropic: Native messages.countTokens() API
• Gemini: SDK’s countTokens() method

Falls back to character-based estimation (4 chars/token) if the provider doesn’t support native token counting or if counting fails.

This is useful for:

• Pre-request cost estimation
• Context window management
• Request batching optimization

Parameters

model

string

Model identifier (e.g., “openai:gpt-4”, “anthropic:claude-3-5-sonnet-20241022”)

messages

LLMMessage[]

Array of messages to count tokens for

Returns

Promise<number>

Promise resolving to the estimated input token count

Example

const client = new LLMist();
const messages = [
  { role: 'system', content: 'You are a helpful assistant.' },
  { role: 'user', content: 'Hello!' }
];

const tokenCount = await client.countTokens('openai:gpt-4', messages);
console.log(`Estimated tokens: ${tokenCount}`);

---

createAgent()

createAgent(): AgentBuilder

Defined in: core/client.ts:325

Create agent builder with this client instance. Useful when you want to reuse a configured client.

Returns

AgentBuilder

AgentBuilder instance using this client

Example

const client = new LLMist({ ... });

const agent = client.createAgent()
  .withModel("sonnet")
  .ask("Hello");

---

stream()

stream(options): LLMStream

Defined in: core/client.ts:142

Parameters

options

LLMGenerationOptions

Returns

LLMStream

---

streamText()

streamText(prompt, options?): AsyncGenerator<string>

Defined in: core/client.ts:275

Instance method: Quick streaming using this client instance.

Parameters

prompt

string

User prompt

options?

TextGenerationOptions

Optional configuration

Returns

AsyncGenerator<string>

Async generator yielding text chunks

---

complete()

static complete(prompt, options?): Promise<string>

Defined in: core/client.ts:224

Quick completion - returns final text response. Convenient for simple queries without needing agent setup.

Parameters

prompt

string

User prompt

options?

TextGenerationOptions

Optional configuration

Returns

Promise<string>

Complete text response

Example

const answer = await LLMist.complete("What is 2+2?");
console.log(answer); // "4" or "2+2 equals 4"

const answer = await LLMist.complete("Tell me a joke", {
  model: "sonnet",
  temperature: 0.9
});

---

createAgent()

static createAgent(): AgentBuilder

Defined in: core/client.ts:306

Create a fluent agent builder. Provides a chainable API for configuring and creating agents.

Returns

AgentBuilder

AgentBuilder instance for chaining

Examples

const agent = LLMist.createAgent()
  .withModel("sonnet")
  .withSystem("You are a helpful assistant")
  .withGadgets(Calculator, Weather)
  .ask("What's the weather in Paris?");

for await (const event of agent.run()) {
  // handle events
}

// Quick one-liner for simple queries
const answer = await LLMist.createAgent()
  .withModel("gpt4-mini")
  .askAndCollect("What is 2+2?");

---

stream()

static stream(prompt, options?): AsyncGenerator<string>

Defined in: core/client.ts:252

Quick streaming - returns async generator of text chunks. Convenient for streaming responses without needing agent setup.

Parameters

prompt

string

User prompt

options?

TextGenerationOptions

Optional configuration

Returns

AsyncGenerator<string>

Async generator yielding text chunks

Example

for await (const chunk of LLMist.stream("Tell me a story")) {
  process.stdout.write(chunk);
}

// With options
for await (const chunk of LLMist.stream("Generate code", {
  model: "gpt4",
  systemPrompt: "You are a coding assistant"
})) {
  process.stdout.write(chunk);
}