The unified surface. - Layer X1

Base URL

All endpoints below are relative to:

https://api.layerx1.com

Authentication is one key across the whole surface — see Authentication for which header each dialect expects.

Endpoints

Endpoint	What it does
`POST` `/v1/chat/completions`	OpenAI Chat Completions dialect — messages in, completion out. SSE streaming.
`POST` `/v1/messages`	Anthropic Messages dialect — content blocks, tools, vision + PDFs. SSE streaming.
`POST` `/v1/responses`	OpenAI Responses dialect — the surface Codex CLI speaks. SSE streaming.
`POST` `/v1/completions`	Legacy text completions for older clients — prompt in, text out.
`POST` `/v1/embeddings`	Seven embedding models, 384 to 2048 dimensions — memory and retrieval next to chat.
`POST` `/v1/messages/count_tokens`	Token estimate for a Messages request body, for context budgeting.
`POST` `/v1/messages/batches`	Async batches — up to 1,000 requests per submission, results as JSONL.
`GET` `/v1/models`	The self-describing model catalog: ids, capabilities, limits, list pricing.

GET /v1/models/{id} returns a single catalog entry. Batches also exist in an OpenAI flavor at /v1/batches with the same lifecycle — see Batches.

The agent-native surface

The endpoints above are one turn each. These are for work that outlives a request — a job you submit, follow, and inspect afterwards.

Endpoint	What it does
`POST` `/v1/agent-runs`	Create a durable run. With a model and input, the gateway executes it and you follow the event stream.
`GET` `/v1/agent-runs/{id}/events`	SSE tail of a run’s event log.
`POST` `/v1/states`	Durable, append-only state handles for an agent’s working memory.
`POST` `/v1/artifacts`	Store a large result once, then query slices of it instead of re-reading it into the prompt.

Also on the surface

Endpoint	What it does
`GET` `/v1/responses/{id}`	Replay a stored Responses envelope.
`GET` `/healthz`	Unauthenticated liveness probe. Returns `ok`.

Conventions

Errors are standard HTTP status codes with a JSON body shaped for the protocol you called (Errors). Responses carry a small, fixed set of x1- and x-ratelimit- headers (Headers). Streaming is Server-Sent Events on every chat endpoint (Streaming).