Choose an Endpoint

Compatible gateways usually expose several protocol entry points. Tools differ in what they support, so choose the endpoint first and then configure the matching Base URL.

This page is the authority for Base URLs

An API node is usually the service root, for example https://gateway.example.com. A Base URL is the value a specific tool expects in its configuration field. They are not always the same.

Use the table on this page as the authoritative Base URL convention for the tool guides in this documentation. If your service requires something different, follow that service's instructions.

Common endpoints

Endpoint	Path	Common authentication header	Typical clients
OpenAI Chat Completions	/v1/chat/completions	Authorization: Bearer <key>	Most OpenAI-compatible tools, OpenCode's openai-compatible provider, Copilot CLI
OpenAI Responses	/v1/responses	Authorization: Bearer <key>	Codex, OpenCode's openai provider, GitHub Copilot CLI
Anthropic Messages	/v1/messages	x-api-key: <key>; some compatible gateways also accept Authorization: Bearer <key>	Claude Code, GitHub Copilot CLI, Anthropic SDKs

These paths are common conventions relative to the gateway root. A compatible service may implement a subset, and actual availability still depends on the gateway, group, model and key permissions.

Base URL reference

The examples below assume the API node is https://gateway.example.com with no trailing /. "Adds /v1 automatically" means whether the tool supplies the /v1 prefix itself; it does not mean the tool supplies the final endpoint path.

Tool	Configuration field or method	Value to enter	Adds /v1 automatically?	Final request path
Codex	model_providers.<id>.base_url in ~/.codex/config.toml	https://gateway.example.com/v1	No; the value must include /v1, and the tool appends /responses	https://gateway.example.com/v1/responses
Claude Code	ANTHROPIC_BASE_URL in settings.json or the environment	https://gateway.example.com	Yes; the tool appends /v1/messages	https://gateway.example.com/v1/messages
OpenCode	provider.<id>.options.baseURL in opencode.json	https://gateway.example.com/v1	No; the value must include /v1, and the AI SDK appends the endpoint path	With @ai-sdk/openai, /v1/responses; with @ai-sdk/openai-compatible, /v1/chat/completions
GitHub Copilot CLI	COPILOT_PROVIDER_BASE_URL in OpenAI-compatible mode	https://gateway.example.com/v1	No; the value must include /v1, and the tool appends the path selected by the wire API	/v1/chat/completions when COPILOT_PROVIDER_WIRE_API=completions; /v1/responses when set to responses

Rules

1. Separate the API node from the Base URL. The node shown in a console usually has no /v1; the value entered into a tool depends on that tool's requirements.
2. Codex must use a Base URL containing /v1. Do not enter https://gateway.example.com directly in base_url, and do not enter /v1/responses.
3. Claude Code uses the domain only. ANTHROPIC_BASE_URL must not include /v1, /v1/messages or a trailing slash.
4. OpenCode and Copilot CLI use a Base URL containing /v1 in OpenAI-compatible mode. For Anthropic provider types, follow the relevant tool page and provider instructions instead of copying the OpenAI example.
5. Do not add a trailing /. It can produce //v1 or otherwise fail to match the service's route.
6. Do not repeat the endpoint path in the Base URL. If the final URL contains /v1/v1/... or /v1/responses/responses, check this first.

Minimal verification requests

The examples below confirm that at least one combination of key, Base URL, model and protocol works. Replace API_KEY, MODEL_ID and the domain. These use bash syntax; on Windows PowerShell, prefer curl.exe and use PowerShell-compatible quoting for the JSON body.

Chat Completions

curl https://gateway.example.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "messages": [{"role": "user", "content": "Reply with OK"}],
    "max_tokens": 16
  }'

Responses

curl https://gateway.example.com/v1/responses \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "input": "Reply with OK",
    "max_output_tokens": 16
  }'

Anthropic Messages

curl https://gateway.example.com/v1/messages \
  -H "x-api-key: $API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "max_tokens": 16,
    "messages": [{"role": "user", "content": "Reply with OK"}]
  }'

If the gateway accepts only Bearer authentication, replace x-api-key in the Anthropic example with Authorization: Bearer $API_KEY. Conversely, if an OpenAI-style endpoint on a compatible service requires x-api-key, follow that service. Some compatible implementations may ignore anthropic-version.

A successful minimal request only proves that this key, model and endpoint work together. It does not prove that the tool-call, reasoning or streaming capabilities required by a coding tool will also work.

Protocol compatibility matrix

Endpoint	Common authentication	Request shape	Best fit	Not guaranteed
Chat Completions	Authorization: Bearer	A messages array and a mature parameter set	Most OpenAI-compatible clients and chat	Responses-only parameters, or every tool's requirements
Responses	Authorization: Bearer	input, max_output_tokens and related fields	Codex, OpenCode's OpenAI provider, Copilot CLI in responses mode	Every compatible gateway implements it; changing one parameter is not a workaround
Anthropic Messages	Usually x-api-key; compatible gateways may also accept Bearer	Top-level system, max_tokens, content blocks and Anthropic semantics	Claude Code, Anthropic SDKs, Copilot CLI's Anthropic provider	OpenAI parameters are not fully equivalent; unsupported fields may be ignored or rejected

For the same key and model:

• Whether one key can call several endpoints depends on the gateway. Some services allow every endpoint; others restrict by protocol, group or key permissions.
• A model appearing in /v1/models does not mean it supports every endpoint. Confirm that the model has a usable channel in the current group and protocol.
• Protocol parameters are not guaranteed to be equivalent. A tool may send a request successfully while the gateway ignores fields, changes tool-call behavior, changes streaming events or returns a different error shape. Trust the actual response and service logs.

Which one

• Codex: use the Responses endpoint and set base_url to https://gateway.example.com/v1.
• Claude Code: use Anthropic Messages and set ANTHROPIC_BASE_URL to https://gateway.example.com.
• OpenCode: the default example uses Responses with baseURL set to https://gateway.example.com/v1; if the service supports only Chat Completions, use @ai-sdk/openai-compatible with the same /v1 Base URL.
• GitHub Copilot CLI: use Chat Completions by default and set the OpenAI-compatible Base URL to https://gateway.example.com/v1; set COPILOT_PROVIDER_WIRE_API=responses when Responses is required.

Not sure which one

Check the tool guide first, then use the Base URL table and minimal requests on this page.

If you have no idea which endpoint your service supports, try this order:

1. First confirm which endpoint the tool requires: Codex only speaks Responses; Claude Code only speaks Anthropic Messages; OpenCode and Copilot CLI can use either Responses or Chat Completions.
2. For tools with a choice, check the provider docs or console for a /v1/responses endpoint. If it exists, use Responses; if not, use Chat Completions (/v1/chat/completions).
3. If you cannot find out, try Chat Completions first - it is the most common compatible endpoint. If you get 404 or 503, switch to Responses and try again.
4. Do not treat Chat Completions as a universal fallback: it will not rescue Codex (needs Responses) or Claude Code (needs Anthropic Messages).

Authentication

• OpenAI Chat Completions and Responses usually use Authorization: Bearer <your API key>.
• Anthropic Messages usually uses x-api-key: <your API key>; some compatible gateways also accept Bearer.
• Follow the actual service. A wrong authentication header or Base URL can produce 401 or 404.