APIs

Every API uses the same keys, models, aliases, limits, history and usage. Pick the one your app speaks.

API	Endpoints	Page
OpenAI Chat Completions	POST /v1/chat/completions	Chat Completions
OpenAI Responses	POST /v1/responses, GET and DELETE /v1/responses/{id}, GET /v1/responses/{id}/input_items	Responses
Anthropic Messages	POST /v1/messages, POST /v1/messages/count_tokens	Messages
Gemini	POST /v1beta/models/{model}:generateContent, :streamGenerateContent, :countTokens	Gemini
OpenAI legacy completions	POST /v1/completions	Completions
Files	/v1/files (OpenAI and Anthropic shapes), POST /upload/v1beta/files and /v1beta/files (Gemini)	Files API
Model list	GET /v1/models, GET /v1/models/{id}, GET /v1beta/models	each API page

Plugboard's own routes live under /plugboard/: usage, status, metrics, the dashboard and key management. /healthz and /readyz answer health checks. See History, usage and monitoring.

List the models a key can use

PLUGBOARD_BASE_URL is http://127.0.0.1:4010/v1 and PLUGBOARD_KEY is your key. Every curl example is in examples/curl/.

list() {
  curl -sS -f "$PLUGBOARD_BASE_URL/models" -H "Authorization: Bearer $PLUGBOARD_KEY"
}

How a request runs

Each request becomes an OpenCode session in the app's own directory. OpenCode's normal request path calls the provider, so its logins, auth plugins and provider fixes all apply.

Plugboard replaces OpenCode's prompt, messages and tools with the caller's. The caller's tools become placeholder tools that do nothing. When the model calls one, plugboard stops the turn and returns the call to the caller. OpenCode never runs a caller's tool. Text streams out as it arrives. Architecture has the details.

Conversation state

Chat Completions, Messages, Gemini and legacy completions carry the whole conversation in every request. Only Responses keeps state on the server: previous_response_id continues a stored answer. See Responses.

Not supported

• conversation on Responses. Use previous_response_id, or send the full input.
• Provider-hosted tools. Web search, code execution, file search, computer use and MCP servers run on the provider's side. Plugboard only hands your function calls back to you.
• Embeddings, audio, image generation and logprobs.
• Batch, fine-tuning, Assistants and vector stores. The Files API stores a file to put it in a prompt, nothing else. Gemini gs:// files are refused too.
• OpenCode V1. Plugboard needs OpenCode V2.

Good to know

• Gemini model ids with / or # do not fit SDK URLs. Give clients an alias. See Gemini.
• State is per server. Limits, the response cache and anchors live in one OpenCode server process. Two servers do not share them, and a restart clears limits and the cache.
• Token counts and costs are estimates. Plugboard has no tokenizer.
• Plugboard is 0.x. A minor release can change an option or a default. Read the changelog before you upgrade.