SitemapSkip to main content
Text size
Dark mode
RecallRunGuides & benchmarks
Learn to build real systemsGuides•Benchmarks•Experiments•Community
Search the site/Free checklist
Read the latest

Guides/OpenAI SDK

OpenAI SDK guide

The OpenAI Python SDK end to end: Responses vs Chat, streaming, structured outputs, tools, embeddings, retries, cost and evals.

  1. 01How the OpenAI API worksUnder every chatbot, agent and 'AI feature' is the same exchange: your code sends an HTTP request containing a model name and some input, and gets back output plus…
  2. 02Setup, keys & your first callSetup takes five minutes: a virtual environment, the openai package, and an API key in an environment variable .
  3. 03Responses API vs Chat CompletionsOpenAI has two APIs for generating text. Chat Completions ( client.chat.completions.create ) is the older, widely copied format: a list of messages in, one message…
  4. 04Models, parameters & tokensWhich model you call decides quality, speed and cost more than any prompt trick.
  5. 05Streaming responsesA model generates output one token at a time. Without streaming you wait for the whole answer, perhaps 8 seconds for a long reply, and see nothing meanwhile.
  6. 06Structured outputs with PydanticMost application code does not want prose. It wants data : a category, an extracted invoice, a list of action items.
  7. 07Function calling: the tool loopA model cannot query your database or call your API. What it can do is ask you to . You describe functions (name, purpose, JSON Schema for the arguments).
  8. 08Built-in toolsFunction tools (part 07) run in your code. Built-in (hosted) tools run on OpenAI's side within the same request: web search , file search over your uploaded…
  9. 09Images, files & audioCurrent models read images and PDFs as naturally as text: a receipt photo, a chart screenshot, a 40-page contract.
  10. 10Embeddings & semantic searchAn embedding is a list of numbers (a vector) that represents the meaning of a piece of text.
  11. 11Async client & concurrencyEach API call spends almost all its time waiting for the model. Classifying 1,000 support tickets one at a time at 1.5 s each takes 25 minutes, and your CPU is idle…
  12. 12Errors, retries & timeoutsNetwork calls fail. Rate limits are hit, servers have bad minutes, requests are malformed, keys expire.
  13. 13Cost & latency optimisationLLM costs scale with tokens × price × requests, and latency scales mostly with output length and model size.
  14. 14Files, vector stores & managed retrievalTo answer questions from your own documents (policies, manuals, contracts), the model needs the relevant passages in its input.
  15. 15Conversation state & memoryThe model is stateless (part 01), so a chat must supply earlier turns on every call.
  16. 16Safety, moderation & data handlingAn LLM feature has a new kind of attack surface: text . Users can type instructions, documents and web pages can contain hidden instructions, and the model's output…
  17. 17The OpenAI Agents SDKIn part 07 you wrote the tool loop by hand. The Agents SDK ( pip install openai-agents ) packages that loop with a few primitives: an Agent (instructions + tools…
  18. 18Testing & evaluating LLM codeLLM features need two different kinds of checking. Tests check your code : does the tool loop pass the right call_id back, handle a 429, parse the schema, refuse an…
  19. 19Realtime & voiceA voice assistant built from three calls (transcribe → think → speak, part 09) works, but each hop adds latency, and conversation feels stilted.
  20. 20Project: retail analyst assistantAsk 'which category earned the most from completed orders?' in plain English and get a correct answer, with the numbers, the SQL that produced them, and caveats…