Add Perplexity as a BYOK provider by jliounis · Pull Request #5094 · microsoft/vscode-copilot-chat

jliounis · 2026-04-30T22:30:42Z

Summary

Adds Perplexity as a Bring-Your-Own-Key (BYOK) language model provider in Copilot Chat. Perplexity exposes an OpenAI-compatible Chat Completions API at https://api.perplexity.ai, so the new provider extends AbstractOpenAICompatibleLMProvider — mirroring the existing xAI and OpenRouter providers.

After merging, users with a Perplexity API key can pick Perplexity from the BYOK provider dropdown and use Sonar models from Copilot Chat.

Changes

src/extension/byok/vscode-node/perplexityProvider.ts — new PerplexityLMProvider extending AbstractOpenAICompatibleLMProvider. Ships a curated list of currently-available models:
- sonar-pro (default)
- sonar
- sonar-reasoning-pro
- sonar-reasoning
src/extension/byok/vscode-node/byokContribution.ts — registers the new provider alongside Anthropic / xAI / OpenRouter / OpenAI / Gemini / Ollama / Azure / customoai.
package.json — adds the "perplexity" vendor entry under languageModelChatProviders with the standard apiKey configuration field.

Attribution header

The provider sends X-Pplx-Integration: vscode-copilot/<package-version> on every outgoing request (via BYOKModelCapabilities.requestHeaders, picked up by OpenAIEndpoint.getExtraHeaders) so Perplexity can attribute traffic to this integration. Set centrally in perplexityProvider.ts — not sprinkled across call sites.

Branding & defaults

Default model: sonar-pro
Branding: Perplexity in user-facing UI/docs (never "Sonar API")
Base URL: https://api.perplexity.ai
Auth: standard Authorization: Bearer <key> (BYOK key field)

Links

Perplexity docs: https://docs.perplexity.ai
Model cards: https://docs.perplexity.ai/models/model-cards

CLA

I'll sign the Microsoft CLA before merge — no blockers expected.

Test plan

Build the extension
Configure a Perplexity API key under the new "Perplexity" provider in BYOK settings
Verify sonar-pro, sonar, sonar-reasoning-pro, and sonar-reasoning show up in the model picker
Send a chat message and confirm streaming completion works
Confirm requests include X-Pplx-Integration: vscode-copilot/<version> header

Wires up Perplexity as a Bring-Your-Own-Key (BYOK) language model provider in Copilot Chat. Perplexity exposes an OpenAI-compatible chat completions API at https://api.perplexity.ai, so the new provider extends AbstractOpenAICompatibleLMProvider, similar to the existing xAI and OpenRouter providers. - Adds src/extension/byok/vscode-node/perplexityProvider.ts with a curated list of Sonar models (sonar-pro, sonar, sonar-reasoning-pro, sonar-reasoning). sonar-pro is the default. - Registers the provider in byokContribution.ts. - Adds the "perplexity" vendor entry to languageModelChatProviders in package.json so the provider appears in the BYOK dropdown with an API key field. - Sends an X-Pplx-Integration: vscode-copilot/<package-version> header on outgoing requests via BYOKModelCapabilities.requestHeaders so Perplexity can attribute traffic to this integration. Docs: https://docs.perplexity.ai

Copilot

Pull request overview

Adds Perplexity as a new BYOK (Bring-Your-Own-Key) language model provider, integrating it into Copilot Chat’s existing OpenAI-compatible BYOK provider framework and surfacing it in VS Code’s BYOK provider configuration UI.

Changes:

Introduces PerplexityLMProvider (OpenAI-compatible) with a curated set of Sonar models and an X-Pplx-Integration attribution header.
Registers Perplexity in the BYOK contribution so it becomes an available provider at runtime.
Adds a new perplexity vendor entry in package.json to expose API key configuration in the BYOK provider dropdown/settings.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
src/extension/byok/vscode-node/perplexityProvider.ts	Adds the new Perplexity BYOK provider, model metadata, and integration header wiring.
src/extension/byok/vscode-node/byokContribution.ts	Registers Perplexity provider alongside other BYOK providers.
package.json	Adds Perplexity as a configurable `languageModelChatProviders` vendor with an `apiKey` field.

Copilot · 2026-04-30T22:35:47Z

+	}
+
+	protected getModelsBaseUrl(): string | undefined {
+		return 'https://api.perplexity.ai';


getModelsBaseUrl() returns a base URL, which causes AbstractOpenAICompatibleLMProvider to always call ${baseUrl}/models for discovery. That contradicts the comment/PR description that Perplexity doesn't have a stable /models endpoint, and it also means the curated PERPLEXITY_KNOWN_MODELS list will never be shown unless the discovery call succeeds and returns those IDs. Consider overriding getAllModels (like CustomOAI does) to return the curated/merged known-model list with url set to the Perplexity base URL, optionally attempting discovery as a best-effort fallback rather than a hard dependency.

Suggested change

return 'https://api.perplexity.ai';

// Perplexity does not provide a stable /models discovery endpoint for this provider.

// Rely on the curated known-model list instead of enabling generic discovery.

return undefined;

Copilot · 2026-04-30T22:35:47Z

+			merged[id] = { ...caps, requestHeaders: { ...integrationHeader, ...(caps.requestHeaders ?? {}) } };
+		}
+		if (remote) {
+			for (const [id, caps] of Object.entries(remote)) {
+				merged[id] = { ...caps, requestHeaders: { ...integrationHeader, ...(caps.requestHeaders ?? {}) } };


The merge order for requestHeaders allows an existing caps.requestHeaders entry to override X-Pplx-Integration, which can accidentally drop or change the attribution header. If the intent is to guarantee the header on every request, apply the integration header last (so it wins) when merging curated and remote model capabilities.

Suggested change

merged[id] = { ...caps, requestHeaders: { ...integrationHeader, ...(caps.requestHeaders ?? {}) } };

}

if (remote) {

for (const [id, caps] of Object.entries(remote)) {

merged[id] = { ...caps, requestHeaders: { ...integrationHeader, ...(caps.requestHeaders ?? {}) } };

merged[id] = { ...caps, requestHeaders: { ...(caps.requestHeaders ?? {}), ...integrationHeader } };

}

if (remote) {

for (const [id, caps] of Object.entries(remote)) {

merged[id] = { ...caps, requestHeaders: { ...(caps.requestHeaders ?? {}), ...integrationHeader } };

Copilot · 2026-04-30T22:35:48Z

+export class PerplexityLMProvider extends AbstractOpenAICompatibleLMProvider {
+
+	public static readonly providerName = 'Perplexity';
+
+	constructor(


A new BYOK provider is introduced but there are no accompanying unit tests under src/extension/byok/vscode-node/test/ (other providers like Ollama/Gemini/Azure have coverage). Adding a Perplexity provider spec would help lock in: the curated model list shown in the picker and that X-Pplx-Integration is present in each model's requestHeaders.

Replace Sonar chat-completions models with the Perplexity Agent API lineup (OpenAI, Anthropic, Google) at the OpenAI-Responses-compatible base URL https://api.perplexity.ai/v1. Also override getAllModels to return the curated list directly (Perplexity does not expose a stable /models discovery endpoint), and flip the requestHeaders merge order so X-Pplx-Integration is never overridden by per-model requestHeaders.

…header Three cases: - getAllModels returns the curated Agent API list with the correct base URL and asserts no sonar references remain. - every model carries the X-Pplx-Integration header in merged capabilities. - per-model requestHeaders cannot override the integration header.

jliounis · 2026-05-07T12:33:36Z

@microsoft-github-policy-service agree

jliounis · 2026-05-07T12:33:55Z

Thanks for the review @copilot-pull-request-reviewer. Pushed two commits addressing all three points and refactoring the provider to expose the Perplexity Agent API instead of Sonar models:

Refactor: Sonar → Agent API

Base URL is now https://api.perplexity.ai/v1 (Agent API; OpenAI-Responses-compatible at /v1/responses and /v1/agent).
Curated list now exposes the Agent API model lineup (OpenAI, Anthropic, Google) under one Perplexity API key, plus access to presets like pro-search.
No Sonar references remain.

1. getModelsBaseUrl → getAllModels override (16b00ef04d2ade9f20da0a754b80c1020f0d6fe9)
Override getAllModels to return the curated Agent API list directly with url set to the Agent API base — AbstractOpenAICompatibleLMProvider's default would call ${baseUrl}/models, which Perplexity does not expose stably. No discovery fallback (intentional — would just slow the picker).

2. Header merge order (16b00ef04d2ade9f20da0a754b80c1020f0d6fe9)
Flipped the spread order so X-Pplx-Integration always wins over any model-level requestHeaders. Applied in both the curated and remote-merge loops.

3. Unit tests (61bc734812e2514c5a3d348b7ce94a3dae268ed2)
Added src/extension/byok/vscode-node/test/perplexityProvider.spec.ts with three cases: curated list shape (asserts no sonar substrings), integration header presence, and header-precedence (per-model overrides cannot drop X-Pplx-Integration).

Also signed the CLA in a separate comment. Let me know if you'd like any adjustments to the curated model list.

jliounis · 2026-05-17T10:31:29Z

CLA is signed. The May 7 push addresses all three Copilot reviewer points (curated model list via getAllModels override, integration header merge order, BYOK provider tests). Could a maintainer authorize Community PR Approvals and take a look? Happy to address any further review feedback.

alexdima · 2026-05-20T10:31:11Z

Thanks for the contribution! This repository has been archived because the project has moved into the main VS Code repository.

Could you please reopen/recreate this PR against:
https://github.com/microsoft/vscode/tree/main/extensions/copilot

We’ll continue reviewing contributions there. Thanks!

Copilot AI review requested due to automatic review settings April 30, 2026 22:30

Copilot started reviewing on behalf of jliounis April 30, 2026 22:31 View session

Copilot AI reviewed Apr 30, 2026

View reviewed changes

jliounis added 2 commits May 7, 2026 12:33

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add Perplexity as a BYOK provider#5094

Add Perplexity as a BYOK provider#5094
jliounis wants to merge 3 commits into
microsoft:mainfrom
jliounis:add-perplexity-byok-provider

jliounis commented Apr 30, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 30, 2026

Uh oh!

Copilot AI Apr 30, 2026

Uh oh!

Copilot AI Apr 30, 2026

Uh oh!

jliounis commented May 7, 2026

Uh oh!

jliounis commented May 7, 2026

Uh oh!

jliounis commented May 17, 2026

Uh oh!

alexdima commented May 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

-		return 'https://api.perplexity.ai';
+		// Perplexity does not provide a stable /models discovery endpoint for this provider.
+		// Rely on the curated known-model list instead of enabling generic discovery.
+		return undefined;

Conversation

jliounis commented Apr 30, 2026

Summary

Changes

Attribution header

Branding & defaults

Links

CLA

Test plan

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

jliounis commented May 7, 2026

Uh oh!

jliounis commented May 7, 2026

Uh oh!

jliounis commented May 17, 2026

Uh oh!

alexdima commented May 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants