Coding agents

Coding agents

Choose a client, configure its endpoint, and verify your first request.

Start with Claude Code, Aider, or opencode for a terminal client whose text connection was tested locally. Complete the Quickstart first so you have credits, an API key, and a model ID.

Choose a guide

ToolWire formatValidation on 2026-10-10
Claude CodeMessagesSuccess · text test
AiderChat CompletionsSuccess · text test
opencodeChat CompletionsSuccess · text test
ContinueChat CompletionsSuccess · CLI text test; IDE reference only
OpenAI SDKChat CompletionsSuccess · Python + TypeScript, text + streaming
Anthropic SDKMessagesSuccess · Python + TypeScript, text + streaming
Cline / Roo CodeChat CompletionsReference only
ZedChat CompletionsReference only
CursorModel-dependentReference only · protocol needs verification
OpenAI Codex CLIResponsesNot supported

“Success” means the exact test scope listed in each guide passed against a local gateway with synthetic upstream responses. It does not establish real-provider performance or successful multi-step code editing. “Reference only” means configuration was checked against sources, but no inference session was completed. “Not supported” identifies a confirmed compatibility problem.

Keep the endpoint and model separate

The gateway accepts OpenAI Chat Completions and Anthropic Messages. Choose the protocol your client supports, then copy a canonical model ID from Marketplace. A model's vendor does not require you to use that vendor's SDK.

The base URL shape depends on the client. Claude Code and the Anthropic SDK take https://unused.market; the OpenAI-compatible configurations in these guides take https://unused.market/v1. Follow the individual guide.

Check the result

Ask for a short greeting, then open Analytics → Requests. Search by model or request ID, and check the newest matching row. For an error, keep x-request-id when your client exposes it and read troubleshooting.

On this page