OpenAPI

The REST API describes itself. The same document the API serves is the only description you need to generate a client.

GET https://api.galleyrender.com/openapi.json

It needs no authentication and is served as application/json.

Fetch the spec

curl -sS https://api.galleyrender.com/openapi.json -o galley-openapi.json

This is not a copy kept alongside the docs — it is generated from the routes the API is running, so the endpoints, request bodies and error shapes in it are what the service actually accepts. If this page and the spec ever disagree, the spec wins.

What is in it

Every /v1 endpoint documented in this reference:

Path	Page
POST /v1/render	Render
POST /v1/render/batch	Batch render
GET /v1/renders/:id, GET /v1/renders	Get a render
GET /v1/templates, GET /v1/templates/:ref, DELETE /v1/templates/:name	Templates
POST /v1/templates	Create a template
POST /v1/templates/:name/versions, GET /v1/templates/:name/versions	Versions
POST /v1/templates/:ref/validate	Validate
GET /v1/usage	Usage

Plus the shared error envelope, so a generated client gets typed errors as well as typed responses. The service-to-service onboarding routes under /v1/internal/* are not part of it: they are not public API.

Generating a client

TypeScript — types only

openapi-typescript emits types with no runtime, which pairs well with the dependency-free fetch client on the SDKs page.

Types for Node or the browser

npx openapi-typescript https://api.galleyrender.com/openapi.json -o galley.d.ts

Using them

import type { paths } from "./galley.d.ts";

type RenderBody = paths["/v1/render"]["post"]["requestBody"]["content"]["application/json"];
type Render = paths["/v1/render"]["post"]["responses"]["200"]["content"]["application/json"];

Any language — a full client

openapi-generator produces a complete client for most languages.

A Python client

npx @openapitools/openapi-generator-cli generate \
  -i https://api.galleyrender.com/openapi.json \
  -g python \
  -o ./galley-client

Swap -g python for go, java, csharp, rust, typescript-fetch or any other generator name.

Python — models only

datamodel-code-generator turns the schemas into Pydantic models without generating transport code.

Pydantic models

pip install datamodel-code-generator
datamodel-codegen \
  --url https://api.galleyrender.com/openapi.json \
  --input-file-type openapi \
  --output-model-type pydantic_v2.BaseModel \
  --output galley_models.py

Pin it in CI

The spec moves when the API does. If your build depends on generated code, vendor a copy and diff it, so a change shows up as a pull request rather than as a surprise:

Fail the build when the spec changes

curl -sS https://api.galleyrender.com/openapi.json |
  python3 -m json.tool --sort-keys > openapi.next.json
diff -u openapi.json openapi.next.json

For agent readers

If you are a model rather than a code generator, there are two documents written for you and both are shorter than the spec:

• https://mcp.galleyrender.com/skill.md — the whole product on one page: tools, formats, cost, template language, error table, rules of thumb.
• https://galleyrender.com/llms.txt (also at https://mcp.galleyrender.com/llms.txt) — a map of every canonical URL, with one line each.

And the MCP endpoint itself, https://mcp.galleyrender.com/mcp, exposes the same operations as tools with descriptions and input schemas already attached — no generation step at all. See SDKs and clients.