Lua Requires Chilkat v11.2.0+

Prove Stateful Multi-Turn Conversation

See more AI Examples

This example is to prove that the 2nd request in a conversation is part of a stateful multi-turn conversation. The 2nd request asks for information provided in the 1st request. Lua

local chilkat = require("chilkat")

-- This example assumes the Chilkat API to have been previously unlocked.
-- See Global Unlock Sample for sample code.

local ai = chilkat.Ai.new()

-- The provider can be "openai", "google", "claude", "deepseek", "xai", or "perplexity".
-- Support for additional providers will be added in future versions of Chilkat.
ai.provider = "google"

-- Use your provider's API key.
ai.api_key = "MY_API_KEY"

-- Choose a model.
ai.model = "gemini-3.6-flash"

-- Create a new conversation to be maintained locally in memory.
-- If the conversation is the first to be created, it is also automatically selected.
local systemMsg = "You are a helpful assistant"
local developerMsg = ""
local conversationName = "test_conversation"
ai:new_convo(conversationName, systemMsg, developerMsg)

-- -------------------------------------------------------------------------------------------
-- Our first request will establish a secret that we'll ask for in the 2nd request.
-- Add a text input.
ai:input_add_text("Please remember this: The word I'll ask for later is BlubberFish.  What's the word?")

-- Ask the AI for text output.
local success = ai:ask("text")
if not success then
    print(ai.last_error_text)
    return
end

-- Note: The inputs accumulated through one or more calls to Input* methods are cleared when the Ask method returns with success.

-- Get the text response.
local sbResponse = chilkat.StringBuilder.new()
ai:get_output_text_sb(sbResponse)
print(sbResponse:get_as_string())
print("----")

-- Sample output:

-- The word is **BlubberFish**.

-- -------------------------------------------------------------------------------------------
-- Let's ask for the word. If this is a stateful conversation, the AI should respond correctly.
ai:input_add_text("OK, I'm going to ask you again, what's the word?")
success = ai:ask("text")
if not success then
    print(ai.last_error_text)
    return
end

-- Get the text response.
sbResponse:clear()
ai:get_output_text_sb(sbResponse)
print(sbResponse:get_as_string())

-- Sample output:

-- The word is **BlubberFish**.

-- -------------------------------------------------------------------------------------------
-- Look at the conversation so far...
--
local json = chilkat.JsonObject.new()
ai:export_convo(conversationName, json)
json.emit_compact = false
print(json:emit())

-- Note: The conversation transcript format exported and imported by Chilkat is always the OpenAI format,
-- regardless of the AI provider.  (Chilkat translates the format from OpenAI to whatever is needed by the actual AI provider being used.)

-- {
--   "input": [
--     {
--       "role": "system",
--       "content": [
--         {
--           "type": "input_text",
--           "text": "You are a helpful assistant"
--         }
--       ]
--     },
--     {
--       "role": "user",
--       "content": [
--         {
--           "type": "input_text",
--           "text": "Please remember this: The word I'll ask for later is BlubberFish.  What's the word?"
--         }
--       ]
--     },
--     {
--       "role": "assistant",
--       "content": [
--         {
--           "type": "output_text",
--           "text": "The word is **BlubberFish**."
--         }
--       ]
--     },
--     {
--       "role": "user",
--       "content": [
--         {
--           "type": "input_text",
--           "text": "OK, I'm going to ask you again, what's the word?"
--         }
--       ]
--     },
--     {
--       "role": "assistant",
--       "content": [
--         {
--           "type": "output_text",
--           "text": "The word is **BlubberFish**."
--         }
--       ]
--     }
--   ]
-- }

-- -------------------------------------------------------------------------------------------
-- If we unselect the conversation, so that no conversation is selected,
-- then the same query results in the AI not knowing the word.

ai.selected_convo = ""

ai:input_add_text("OK, I'm going to ask you again, what's the word?")
success = ai:ask("text")
if not success then
    print(ai.last_error_text)
    return
end

-- Get the text response.
sbResponse:clear()
ai:get_output_text_sb(sbResponse)
print(sbResponse:get_as_string())

-- Sample output:

-- I apologize, but I don't have the context of our previous interaction or what "the word" you're referring to might be.
-- Could you please remind me or provide the clue again? I'm ready to help you figure it out!