Set up your agent

Almost every AI coding tool connects to Helmcode with the same two values: the base URL and your API key. What changes from one to the next is where you write them.

Field	Value
Base URL	`https://api.helmcode.com/v1`
API key	yours, from the console — it starts with sk-

Every page in this section has the same shape: what you need, the exact configuration with the file path, how to check that it works, and the known issues.

Pick your tool

Claude Code Terminal via gateway It speaks Anthropic’s format. Either you delegate to OpenCode, or you run a gateway. Codex Terminal One provider of your own in config.toml and you are done. Cursor Editor The option is no longer on every plan. Check before you start. VS Code and Copilot Editor Chat and agent mode, yes. Inline completion stays Copilot’s. Cline VS Code extension OpenAI Compatible provider, four fields. OpenCode Terminal Helmcode is declared as one more provider. Zed Editor The only one that lets you use Helmcode for edit predictions too. Pi Terminal A minimal shell. Declare Helmcode and make it the default. OT Other tools Various Continue, Aider, OpenClaw, Open WebUI and LM Studio.

Yours is not here? If it accepts an OpenAI base URL and an API key, it works. Copy the two values from the table above and paste the model id exactly as it appears in Models.

Exactly as it appears is not a figure of speech: an id we do not serve comes back as a 401 saying your key has no access to that model, not as a not-found. `glm-5.3` and `glm5.3` are a typo apart and only one of them exists, and the error for the other one reads like a plan problem.

Separately from the models, Helmcode exposes its web search as a tool for any MCP-capable agent — including Claude Code, whichever models it is using. That is the MCP server.

Building an agent of your own against the API, rather than connecting one? That is Agents.

Why almost all of them work

Helmcode exposes an OpenAI-compatible API. The routes, the request fields and the shape of the response are the ones these tools already expect, so as far as they are concerned Helmcode is indistinguishable from OpenAI except for the address they call.

The exception is Claude Code, which does not speak that format but Anthropic’s. It is not a credentials problem: they are two different protocols. That is why its page explains the two paths there are.

Which model to use

Almost all of these tools ask you for a model id on top of the base URL and the key. For agent work, the sensible order is:

glm5.3-flash — the default for agent sessions: a 1M window, and it counts against the monthly quota every plan already includes.
deepseek-v4-flash — the other 1M model on that same quota — reach for it when the reasoning has to hold across a long refactor.
qwen3.6 — when the task is bounded and you want the answer now — roughly twice the tokens per second, on a 256K window.

glm5.3 — the 744B one, not the Flash — is an add-on with its own allowance rather than part of the plan, so do not standardize a team on it without checking that your organization has it.

The full list, with specs and capabilities, is in Models.

Warning: Careful with the context you declare. Several of these tools let you type the model’s context window by hand. Set a number higher than the real one and the agent will fill the conversation up to a point where the API starts rejecting requests. Set a lower one and it will only compact earlier than needed. When in doubt, stay short.