TOpenAI / TChatGPT

Source: source/classes/topenai.prg (TOpenAI), source/classes/chatgpt.prg (TChatGPT)

FiveWin provides two classes for integrating with the OpenAI API: TOpenAI (full-featured with vision, streaming, and image support) and TChatGPT (simpler wrapper for basic chat completions). Both use Harbour's hb_curl library for HTTPS communication.

Architecture

TOpenAI Class

DATA Members

Methods

TChatGPT Class

TChatGPT is a simpler alternative with fewer features, suitable for basic prompt/response interactions.

DATA Members

Methods

Example: Basic Prompt/Response (TOpenAI)

#include "FiveWin.ch"

function Main()

   local oAI, cAnswer

   // API key from environment variable OPENAI_API_KEY
   oAI := TOpenAI():New()

   // Or with explicit key and model:
   // oAI := TOpenAI():New( "sk-...", "gpt-4o" )

   oAI:Send( "What is the capital of France? Answer in one word." )

   if oAI:nError == 0 .and. oAI:nHttpCode == 200
      cAnswer := oAI:GetValue()
      MsgInfo( cAnswer )   // "Paris"
   else
      MsgStop( "Error: " + oAI:cResponse )
   endif

   oAI:End()

return nil

Example: Streaming Response

#include "FiveWin.ch"

function Main()

   local oAI, cBuffer := ""

   oAI := TOpenAI():New()
   oAI:cModel := "gpt-4o"

   // Stream callback receives chunks of data as they arrive
   oAI:SendStream( "Write a short poem about coding.", ;
      { |cChunk| cBuffer += cChunk, QOut( cChunk ) } )

   if oAI:nError == 0
      MsgInfo( "Complete response:" + CRLF + cBuffer )
   endif

   oAI:End()

return nil

Example: Vision (Analyze an Image)

#include "FiveWin.ch"

function Main()

   local oAI

   oAI := TOpenAI():New( , "gpt-4o" )

   // Send a local image file
   oAI:SendImage( "C:\photos\chart.png", "Describe what you see in this image." )

   if oAI:nError == 0
      MsgInfo( oAI:GetValue() )
   endif

   // Send an image by URL
   oAI:SendImageURL( ;
      "https://example.com/photo.jpg", ;
      "What objects are in this photo?" )

   if oAI:nError == 0
      MsgInfo( oAI:GetValue() )
   endif

   oAI:End()

return nil

Example: TChatGPT (Simple Wrapper)

#include "FiveWin.ch"

function Main()

   local oChat

   oChat := TChatGPT():New( "sk-your-api-key-here" )

   // Set the prompt and send
   oChat:cPrompt := "Translate 'Hello, how are you?' to Spanish"
   oChat:Send()

   if oChat:nError == 0
      // Navigate the JSON response
      ? oChat:GetValue( "choices", "message", "content" )
      // Output: "Hola, como estas?"
   endif

   // Send another prompt (reset first)
   oChat:Reset()
   oChat:cPrompt := "Now translate it to French"
   oChat:Send()

   if oChat:nError == 0
      ? oChat:GetValue( "choices", "message", "content" )
   endif

   oChat:End()

return nil

Example: Adjusting Temperature

#include "FiveWin.ch"

function Main()

   local oAI

   oAI := TOpenAI():New()

   // Low temperature = more deterministic
   oAI:nTemperature := 0.1
   oAI:Send( "What is 2 + 2?" )
   ? "Precise:", oAI:GetValue()

   // High temperature = more creative
   oAI:nTemperature := 1.5
   oAI:Send( "Write a creative product name for a new coffee brand" )
   ? "Creative:", oAI:GetValue()

   oAI:End()

return nil

Supported Models

Tips

• Store your API key in the OPENAI_API_KEY environment variable rather than hardcoding it.
• TOpenAI:New() automatically reads from the environment if no key is passed.
• The hb_curl library must be linked. Use #include "hbcurl.ch" for constants.
• SSL verification is disabled by default (HB_CURLOPT_SSL_VERIFYPEER = .F.) for compatibility.
• Use SendStream() for long responses to get real-time output instead of waiting for the complete response.
• The GetValue() method handles JSON parsing internally. For TOpenAI it returns choices[1].message.content. For TChatGPT, pass the key path as arguments.

TOLlama (Local AI)

Source: source/classes/tollama.prg

Local AI Connects to a local Ollama server. No API key needed - runs AI models on your own machine.

local oAI := TOLlama():New( "llama3" )

oAI:Send( "Explain what FiveWin is in one paragraph" )
MsgInfo( oAI:GetValue() )

// Streaming response
oAI:SendStream( "Write a Harbour function", ;
   {| cChunk | QOut( cChunk ) } )

oAI:End()

TDeepSeek

Source: source/classes/tdeepseek.prg

Cloud AI DeepSeek API client. Known for strong coding and reasoning capabilities.

local oAI := TDeepSeek():New( "your-api-key" )
oAI:Send( "Write a Harbour function to calculate factorial" )
MsgInfo( oAI:GetValue() )
oAI:End()

TGemini (Google)

Source: source/classes/tgemini.prg

Cloud AI Google Gemini API client. Supports text, images, and file uploads.

local oAI := TGemini():New( "your-google-api-key" )
oAI:Send( "Analyze this image", "photo.jpg" )
MsgInfo( oAI:GetValue() )
oAI:End()

TGrok (xAI)

Source: source/classes/tgrok.prg

Cloud AI xAI Grok API client. Supports text, images, and streaming.

local oAI := TGrok():New( "your-xai-key" )
oAI:Send( "What are the benefits of Harbour language?" )
MsgInfo( oAI:GetValue() )
oAI:End()

TKimi (Moonshot)

Source: source/classes/tkimi.prg

Cloud AI Moonshot Kimi API client. Strong multilingual capabilities.

local oAI := TKimi():New( "your-moonshot-key" )
oAI:Send( "Translate this to Chinese: Hello World" )
MsgInfo( oAI:GetValue() )
oAI:End()

AI Classes Comparison

Transformer (Neural Network in Harbour)

Source: source/classes/transformer.prg

Advanced A full Transformer neural network implemented in pure Harbour. Supports training, forward/backward propagation, multi-head attention, and text prediction.

Example: Next Word Prediction (transf1.prg)

local vocab := { "hola", "mundo", "adios", "gato", "perro" }
local oT := Transformer():New( 1, 6, 2, Len(vocab), 8,, vocab )

// Train on sequences
for e = 1 to 60
   for i = 1 to Len(train_src)
      output := oT:Forward( train_src[i], train_pos[i] )
      oT:Backward( output, train_tgt[i], train_src[i], train_pos[i] )
   next
next

// Predict next word
output := oT:Forward( { 1, 2 }, { 1, 2 } )  // "hola mundo" -> ?

Example: Sentiment Classification (transf2.prg)

// Classify phrases as Positive or Negative
local vocab := { "buen", "dia", "malo", "trabajo", "feliz", "triste", "excelente", "horrible" }
local oT := TransformerClassifier():New( 1, 6, 2, 2, vocab )

// Train: "buen dia" -> Positive, "malo trabajo" -> Negative
// After training, classify new phrases

TNeuralNetwork (Classic Neural Network)

Source: source/classes/tneuralnet.prg

Classic feedforward neural network with backpropagation. Good for simple pattern recognition.

// XOR training example
local oNN := TNeuralNetwork():New( 2, { 2 }, 1 )  // 2 inputs, 1 hidden layer with 2 neurons, 1 output

for n = 1 to 30000
   oNN:Learn( { 0, 0 }, { 0 }, .F. )
   oNN:Learn( { 1, 0 }, { 1 }, .F. )
   oNN:Learn( { 0, 1 }, { 1 }, .F. )
   oNN:Learn( { 1, 1 }, { 0 }, .F. )
next

? oNN:Predict( { 1, 0 } )  // -> ~1.0
? oNN:Predict( { 1, 1 } )  // -> ~0.0

TEmbeddings (Semantic Similarity)

Source: source/classes/tembeddings.prg

Text embeddings for semantic similarity comparison. Uses HuggingFace or OpenAI APIs.

local oE1 := TEmbeddings():New()
local oE2 := TEmbeddings():New()
local oE3 := TEmbeddings():New()

oE1:GetEmbeddings( "I feel very well" )
oE2:GetEmbeddings( "How are you?" )
oE3:GetEmbeddings( "I saw Peter" )

? oE1 - oE2   // High similarity (both about feelings)
? oE1 - oE3   // Low similarity (different topics)

AI Agents (TOLlama + TAgent)

AI agents combine LLMs with tools (functions) that the AI can call. The AI decides which tool to use based on the user's request.

local oLlama := TOLlama():New()
local oAgent

// Agent for date/time
oAgent := TAgent():New( "datetime", { ;
   { "get_time", {|hParams| GetCurrentTime(hParams) } }, ;
   { "get_date", {|hParams| GetDate(hParams) } } } )
AAdd( oLlama:aAgents, oAgent )

// Agent for filesystem
oAgent := TAgent():New( "filesystem", { ;
   { "create_folder", {|hParams| CreateFolder(hParams) } }, ;
   { "create_file",   {|hParams| CreateFile(hParams) } } } )
AAdd( oLlama:aAgents, oAgent )

// AI decides which tool to use
oLlama:Send( "What time is it?" )       // -> calls get_time
oLlama:Send( "Create a folder test" )   // -> calls create_folder

AI Samples Reference

All AI samples are in the samples/ai/ folder:

See Also

• Classes Alphabetical
• Functions