Zukko Blog
Guides to Claude Code, Cursor, Codex and the Claude and GPT APIs: setup, token savings and troubleshooting.
Getting strictly structured JSON out of the model
3 min · · API
Asking the model to "reply in JSON" in the system prompt works about half the time. Here is a more reliable approach using tools with a strict schema.
Why Russian text tokenizes differently than English
4 min · · models
The same sentence in Russian and English isn't billed the same — not because of markup, but because the tokenizer splits Cyrillic differently.
Timeouts on long generations: how to avoid them
4 min · · development
A response cut off mid-stream almost always means your own client's timeout, not a token limit. Where it comes from and how to retry without a double charge.
Full Node.js example: official SDKs through our gateway
2 min · · development
The Node builds of the official SDKs configure exactly like their Python counterparts — one baseURL field reroutes the whole request.
Full Python example: official SDKs through our gateway
3 min · · development
You don't need to replace the official SDKs, just repoint them. A working walkthrough of both clients against real endpoints.
OpenAI Agents SDK over our /v1/responses endpoint
4 min · · агенты
The Agents SDK talks to OpenAI by default. Two client settings redirect it through our gateway onto the models in our catalog.
LlamaIndex: indexing data with Claude and GPT
5 min · · development
LlamaIndex splits the answer-synthesis LLM from the embedding model — the gateway only covers the first part. Here's how to set that up honestly.
Vercel AI SDK: connecting a Next.js app to Claude and GPT
5 min · · development
The AI SDK's OpenAI and Anthropic providers build the path on top of baseURL differently — easy to get /v1 wrong in one route handler. Setup and the client-leak trap.
Semantic Kernel: connecting a .NET app to Claude and GPT
5 min · · агенты
SK talks to OpenAI through the official .NET SDK, which honors HttpClient.BaseAddress. Setup, the trailing-slash 404, and double-charged retries.
Bulk request processing without hitting the rate limit
4 min · · development
One key with a spending limit on a batch job is a guaranteed 429 partway through. How to structure bulk processing so that doesn't happen.
Wiring an AI bot into Slack over API
4 min · · development
Slack expects a reply to a command within three seconds, and a model can take longer. How to work around it with response_url, with working Bolt code.
AI tools in the terminal: what exists and when to use what
2 min · · cli
Not everything called an agent is one. Three classes of tools and the jobs they were built for.
VS Code AI extensions: how they differ
2 min · · setup
There are many extensions; they differ in three things: how they connect a model, how they build context, how they ask permission.
Migrating a project from the official API to a gateway
2 min · · migration
Moving to a compatible gateway is two lines of configuration and an hour of verification. Here is exactly what changes.
Your API key leaked — a step-by-step response
2 min · · security
A leaked key means a growing bill. The first steps take a minute, and their order matters more than speed.