7.7.1. OPENAI-01 — Your First Chat

This tutorial introduces the openai module — a pure-daslang client for OpenAI-compatible APIs (OpenAI, Ollama, OpenRouter, Kokoro, and any server that speaks the same REST surface). It builds on dashv for HTTP and daslib/json_boost for serialization.

The companion .das files are in tutorials/dasOPENAI/. They run against a small local mock server (tutorial_openai_server.das) so no live LLM is needed.

7.7.1.1. Setup

Require the section module you need and set options rtti:

options gen2
options rtti          // REQUIRED — see below

require openai/openai_chat

options rtti is mandatory on every dasOPENAI script. Without it the @optional / @rename field annotations are ignored, and requests serialize every field (e.g. "n":0) — which real servers reject.

7.7.1.2. Constructing a client

openai_client takes a base URL (no trailing slash) and an optional API key:

// local server, no key
let client = openai_client("http://localhost:11434/v1")

// real OpenAI, key from the environment
let client = openai_client("https://api.openai.com/v1", get_env("OPENAI_API_KEY"))

7.7.1.3. The chat_text one-liner

chat_text is the simplest entry point: one user turn in, the assistant’s text out (empty string on error).

let reply = chat_text(client, "gpt-4o-mini", "Say hello in one short sentence.")
print("{reply}\n")

7.7.1.4. The full chat() call

chat takes a ChatCompletionRequest and returns a ChatResult. When ok is true, response holds the decoded reply — choices, the assistant message, and token usage:

let req = ChatCompletionRequest(model = "gpt-4o-mini",
    messages <- [ChatMessage(role = "user", content = "What is 2+2?")])
let res = chat(client, req)
if (res.ok) {
    print("answer: {res.response.choices[0].message.content}\n")
    print("tokens: {res.response.usage.total_tokens}\n")
} else {
    print("error [{res.error.kind}/{res.error.status}]: {res.error.message}\n")
}

7.7.1.5. Handling errors

Every call returns a unified OpenAIError rather than crashing. kind is one of "transport" (could not reach the server), "http", "api" (the server returned an error body), or "decode" (the response did not parse).

7.7.1.6. Quick Reference

Function	Description
`openai_client(base_url [, api_key])`	Build a client for any OpenAI-compatible server
`chat_text(client, model, prompt)`	One user turn → assistant text (”” on error)
`chat(client, req)`	Full request → `ChatResult` (ok/response/error)
`res.response.choices[0].message`	The assistant `ChatMessage`
`res.response.usage.total_tokens`	Token accounting
`res.error.kind` / `.status`	Error category and HTTP status