.. | ||
auto | ||
bun-runtime.d.ts | ||
bun-runtime.d.ts.map | ||
bun-runtime.js | ||
bun-runtime.js.map | ||
bun-runtime.mjs | ||
bun-runtime.mjs.map | ||
index.d.ts | ||
index.js | ||
index.mjs | ||
manual-types.d.ts | ||
manual-types.js | ||
manual-types.mjs | ||
MultipartBody.d.ts | ||
MultipartBody.d.ts.map | ||
MultipartBody.js | ||
MultipartBody.js.map | ||
MultipartBody.mjs | ||
MultipartBody.mjs.map | ||
node-runtime.d.ts | ||
node-runtime.d.ts.map | ||
node-runtime.js | ||
node-runtime.js.map | ||
node-runtime.mjs | ||
node-runtime.mjs.map | ||
node-types.d.ts | ||
node-types.js | ||
node-types.mjs | ||
README.md | ||
registry.d.ts | ||
registry.d.ts.map | ||
registry.js | ||
registry.js.map | ||
registry.mjs | ||
registry.mjs.map | ||
web-runtime.d.ts | ||
web-runtime.d.ts.map | ||
web-runtime.js | ||
web-runtime.js.map | ||
web-runtime.mjs | ||
web-runtime.mjs.map | ||
web-types.d.ts | ||
web-types.js | ||
web-types.mjs |
👋 Wondering what everything in here does?
openai
supports a wide variety of runtime environments like Node.js, Deno, Bun, browsers, and various
edge runtimes, as well as both CommonJS (CJS) and EcmaScript Modules (ESM).
To do this, openai
provides shims for either using node-fetch
when in Node (because fetch
is still experimental there) or the global fetch
API built into the environment when not in Node.
It uses conditional exports to
automatically select the correct shims for each environment. However, conditional exports are a fairly new
feature and not supported everywhere. For instance, the TypeScript "moduleResolution": "node"
setting doesn't consult the exports
map, compared to "moduleResolution": "nodeNext"
, which does.
Unfortunately that's still the default setting, and it can result in errors like
getting the wrong raw Response
type from .asResponse()
, for example.
The user can work around these issues by manually importing one of:
import 'openai/shims/node'
import 'openai/shims/web'
All of the code here in _shims
handles selecting the automatic default shims or manual overrides.
How it works - Runtime
Runtime shims get installed by calling setShims
exported by openai/_shims/registry
.
Manually importing openai/shims/node
or openai/shims/web
, calls setShims
with the respective runtime shims.
All client code imports shims from openai/_shims/index
, which:
- checks if shims have been set manually
- if not, calls
setShims
with the shims fromopenai/_shims/auto/runtime
- re-exports the installed shims from
openai/_shims/registry
.
openai/_shims/auto/runtime
exports web runtime shims.
If the node
export condition is set, the export map replaces it with openai/_shims/auto/runtime-node
.
How it works - Type time
All client code imports shim types from openai/_shims/index
, which selects the manual types from openai/_shims/manual-types
if they have been declared, otherwise it exports the auto types from openai/_shims/auto/types
.
openai/_shims/manual-types
exports an empty namespace.
Manually importing openai/shims/node
or openai/shims/web
merges declarations into this empty namespace, so they get picked up by openai/_shims/index
.
openai/_shims/auto/types
exports web type definitions.
If the node
export condition is set, the export map replaces it with openai/_shims/auto/types-node
, though TS only picks this up if "moduleResolution": "nodenext"
or "moduleResolution": "bundler"
.