Nous.LLM (nous v0.13.3)

View Source

Simple LLM API for direct model calls without agents.

This module provides a lightweight interface for making LLM calls without the full agent machinery. Use this when you need simple text generation, optionally with tools.

Examples

# Simple generation
{:ok, text} = Nous.generate_text("openai:gpt-4", "What is 2+2?")
IO.puts(text) # "4"

# With options
text = Nous.generate_text!("anthropic:claude-haiku-4-5", "Hello",
  system: "You are a pirate",
  temperature: 0.7,
  max_tokens: 500
)

# With tools
{:ok, text} = Nous.generate_text("openai:gpt-4", "What's the weather in Paris?",
  tools: [&MyTools.get_weather/2]
)

# Streaming
{:ok, stream} = Nous.stream_text("openai:gpt-4", "Write a story")
stream |> Stream.each(&IO.write/1) |> Stream.run()

Summary

Types

option()

Functions

generate_text(model, prompt, opts \\ [])

Generate text from a model.

generate_text!(model, prompt, opts \\ [])

Generate text from a model, raising on error.

stream_text(model, prompt, opts \\ [])

Stream text from a model.

Types

option()

@type option() ::
  {:system, String.t()}
  | {:temperature, float()}
  | {:max_tokens, pos_integer()}
  | {:top_p, float()}
  | {:base_url, String.t()}
  | {:api_key, String.t()}
  | {:receive_timeout, non_neg_integer()}
  | {:tools, [function() | Nous.Tool.t()]}
  | {:deps, map()}
  | {:fallback, [String.t() | Nous.Model.t()]}

Functions

generate_text(model, prompt, opts \\ [])

@spec generate_text(String.t() | Nous.Model.t(), String.t(), [option()]) ::
  {:ok, String.t()} | {:error, term()}

Generate text from a model.

Returns {:ok, text} on success, {:error, reason} on failure.

If tools are provided and the model calls them, they will be executed automatically and the conversation will continue until the model returns a text response.

Parameters

  • model - Model string ("provider:model-name") or %Model{} struct
  • prompt - The user prompt
  • opts - Options (see below)

Options

  • :system - System prompt
  • :temperature - Sampling temperature (0.0 to 2.0)
  • :max_tokens - Maximum tokens to generate
  • :top_p - Nucleus sampling parameter
  • :base_url - Override API base URL
  • :api_key - Override API key
  • :receive_timeout - HTTP receive timeout in milliseconds (default varies by provider)
  • :tools - List of tool functions or Tool structs
  • :deps - Dependencies to pass to tool functions
  • :fallback - Ordered list of fallback model strings or Model structs to try when the primary model fails with a provider/model error

Examples

{:ok, text} = Nous.LLM.generate_text("openai:gpt-4", "What is 2+2?")

{:ok, text} = Nous.LLM.generate_text("anthropic:claude-haiku-4-5", "Hello",
  system: "You are helpful",
  temperature: 0.7
)

# With tools
{:ok, text} = Nous.LLM.generate_text("openai:gpt-4", "What's the weather?",
  tools: [&MyTools.get_weather/2],
  deps: %{api_key: "..."}
)

generate_text!(model, prompt, opts \\ [])

@spec generate_text!(String.t() | Nous.Model.t(), String.t(), [option()]) ::
  String.t()

Generate text from a model, raising on error.

Same as generate_text/3 but raises Nous.Errors.ModelError on failure.

Examples

text = Nous.LLM.generate_text!("openai:gpt-4", "What is 2+2?")
IO.puts(text) # "4"

stream_text(model, prompt, opts \\ [])

@spec stream_text(String.t() | Nous.Model.t(), String.t(), [option()]) ::
  {:ok, Enumerable.t()} | {:error, term()}

Stream text from a model.

Returns {:ok, stream} where stream yields text chunks as strings.

Parameters

  • model - Model string ("provider:model-name") or %Model{} struct
  • prompt - The user prompt
  • opts - Options (same as generate_text/3)

Examples

{:ok, stream} = Nous.LLM.stream_text("openai:gpt-4", "Write a haiku")
stream |> Stream.each(&IO.write/1) |> Stream.run()