fa_llm 0.2.1

Reusable LLM provider adapters for Fa and related projects. Pure Dart core: OpenAI-compatible, OpenRouter, Ollama, and configuration.

fa_llm

Reusable LLM provider adapters for Fa and related projects.

Pure Dart core: OpenAI-compatible, OpenRouter, Ollama, and configuration.

Usage

import 'package:fa_llm/fa_llm.dart';

final provider = ProviderFactory.create(
  LlmConfig(
    providerName: 'openai',
    apiKey: 'sk-...',
    baseUrl: 'https://api.openai.com/v1/chat/completions',
    model: 'gpt-4o',
  ),
);

Configuration

LlmConfig.fromEnvironment() resolves each value in order: the explicit argument, then the {PROVIDER}_… environment variable, then the same name in a project-root .env file, then the built-in default. PROVIDER is OPENAI (the default), OPENROUTER, or OLLAMA.

Variable	Value
{PROVIDER}_API_KEY	API key
{PROVIDER}_MODEL	Model id
{PROVIDER}_BASE_PATH	Base URL — wins over {PROVIDER}_BASE_URL when both are set
{PROVIDER}_BASE_URL	Base URL (see below)
{PROVIDER}_MAX_TOKENS	Max output tokens (int, optional)
{PROVIDER}_CONTEXT_WINDOW	Context window (int, optional)
{PROVIDER}_TEMPERATURE	Sampling temperature (double, optional)
{PROVIDER}_MAX_TOKENS_PARAM_NAME	max_tokens payload key (default max_completion_tokens)

An origin (http://127.0.0.1:8931) or a versioned base (http://127.0.0.1:8931/v1) is normalized to the full chat-completions endpoint the providers POST to, so a local proxy (recorder, gateway, cost tracker) can be pointed at with its origin alone. Anything else — a full endpoint, a custom gateway path, a URL with a query string — passes through verbatim. The copilot provider's base URL is its API origin and is never rewritten.

Features

• OpenAI-compatible streaming completions
• OpenRouter model routing
• Ollama local inference
• Provider configuration and resolution
• Token counting and context window management