Add fetch-native proxy support (HTTPS_PROXY / HTTP_PROXY / NO_PROXY)

[robgruen]

Summary

Adds optional proxy support for the model endpoints on the native fetch stack, and consolidates the OpenAI factory's optional arguments into a single LanguageModelOptions bag. This supersedes #55 and #73, both written against the old axios-based model.ts (axios has since been removed).

Proxy support

• createLanguageModel(env) auto-detects the standard HTTPS_PROXY / HTTP_PROXY / ALL_PROXY variables and honors NO_PROXY, using undici's EnvHttpProxyAgent.
• createOpenAILanguageModel / createAzureOpenAILanguageModel accept an explicit { proxyUrl }.
• The proxy is supplied as the fetch dispatcher — the only supported proxy hook for Node's native fetch (which is itself undici).
• undici is an optional, lazily-imported dependency (optional peer dep). Non-proxy users are unaffected; a clear "run npm install undici" error is thrown if a proxy is configured without it.

Why undici

Node's built-in fetch is undici and exposes no proxy option other than a dispatcher, which must be an undici agent. https-proxy-agent (used by #55) returns an http.Agent, which native fetch ignores — so undici is the natural choice once the project is on fetch.

API consolidation

LanguageModelOptions now also carries useResponsesApi, replacing the positional useResponsesApi parameter on createOpenAILanguageModel:

// before (unreleased):
createOpenAILanguageModel(apiKey, model, endPoint?, org?, useResponsesApi?, options?)
// after:
createOpenAILanguageModel(apiKey, model, endPoint?, org?, options?)  // { proxyUrl?, useResponsesApi? }

The positional useResponsesApi and the options object were both introduced after the published 0.1.2 (whose signature is createOpenAILanguageModel(apiKey, model, endPoint?, org?)), so this is not a breaking change to any released API. Doing it now — before either ships — avoids a future breaking change.

Version

Bumps typechat to 0.1.3.

Testing

Validated end-to-end with a local mock endpoint + CONNECT-tunneling proxy:

Scenario	Result	Routed through proxy?
no proxy	success	no (direct)
{ proxyUrl } / HTTP_PROXY set	success	yes
empty proxy string ("")	success	no — no agent constructed (fixes #55's crash-on-default)
NO_PROXY matches host	success	no — bypassed
NO_PROXY does not match	success	yes

tsc -p src, tsc -p test, and the model test suite (19 tests) all pass.

Notes

• EnvHttpProxyAgent is currently flagged experimental by undici and prints a one-time warning; the core ProxyAgent / dispatcher path is stable.
• No new required dependencies for existing users.

Opening as a draft for discussion.

[Copilot]

Pull request overview

This PR adds optional HTTP(S) proxy support for the TypeScript language-model factories when using Node’s native fetch (undici), and consolidates OpenAI factory optional parameters into a single LanguageModelOptions object (including useResponsesApi and proxyUrl). It also bumps the TypeScript package version to 0.1.3.

Changes:

• Add proxy discovery from standard env vars (HTTPS_PROXY / HTTP_PROXY / ALL_PROXY, honoring NO_PROXY) and support an explicit proxyUrl option via undici dispatchers.
• Replace the positional useResponsesApi parameter with options.useResponsesApi on createOpenAILanguageModel.
• Update package version and dependencies/docs/tests accordingly.

Show a summary per file

File	Description
typescript/src/model.ts	Introduces LanguageModelOptions, proxy resolution (env + explicit URL), and threads an undici dispatcher into fetch calls.
typescript/tests/model.test.mjs	Updates a test to use the new options.useResponsesApi API.
typescript/package.json	Bumps version to 0.1.3 and adds undici as an optional peer dependency (and dev dependency for development/tests).
typescript/package-lock.json	Lockfile updates for version bump and dev dependency changes.
typescript/examples/README.md	Documents proxy usage via env vars and direct factory options.

Review details

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Files not reviewed (1)

• typescript/package-lock.json: Generated file

• Files reviewed: 4/5 changed files
• Comments generated: 1
• Review effort level: Low