feat(providers): add Perplexity as a supported model provider

[jliounis]

Summary

Adds Perplexity as a supported model provider in crates/goose/src/providers/, mirroring the layout of the existing xAI provider (xai.rs).

Perplexity's chat completions API is OpenAI-compatible (https://api.perplexity.ai), so the provider is a thin ProviderDef that returns an OpenAiCompatibleProvider configured with bearer-token auth. Its chat models ship with built-in real-time web search grounding, which is a natural fit for a developer agent that frequently needs to ground answers in current information.

What's added

• crates/goose/src/providers/perplexity.rs (new) — PerplexityProvider:
  • Base URL: https://api.perplexity.ai (configurable via PERPLEXITY_HOST)
  • Auth: PERPLEXITY_API_KEY (canonical) → PPLX_API_KEY (alias used by Perplexity's SDKs) → error
  • Default model: sonar-pro. Known models: sonar, sonar-pro, sonar-reasoning, sonar-reasoning-pro (users can override GOOSE_MODEL to point at any other model the API accepts)
  • Setup steps wired up for the onboarding wizard.
• crates/goose/src/providers/mod.rs — register pub mod perplexity;
• crates/goose/src/providers/init.rs — register PerplexityProvider in the registry (default-off, matching xAI/Venice)
• crates/goose/tests/providers.rs — test_perplexity_provider smoke test entry following the same ProviderTestConfig::with_llm_provider shape as test_xai_provider
• documentation/docs/getting-started/providers.md — Perplexity row added to the provider matrix

Tests

5 unit tests in crates/goose/src/providers/perplexity.rs (alongside provider source, matching the convention used by venice.rs):

1. test_metadata_structure — checks name, display_name, default_model, model_doc_link, and both ConfigKeys (PERPLEXITY_API_KEY required+secret+primary, PERPLEXITY_HOST with default).
2. test_known_models_non_empty — default model is included in the known list.
3. test_setup_steps_present — onboarding steps reference PERPLEXITY_API_KEY.
4. test_default_model_is_known.
5. test_doc_url_points_to_perplexity_docs.

Plus 1 integration test entry (test_perplexity_provider) in crates/goose/tests/providers.rs, which the existing test harness skips when PERPLEXITY_API_KEY isn't set (same pattern as every other LLM provider here).

Validation

cargo test -p goose providers::perplexity
cargo clippy --all-targets -- -D warnings
cargo fmt --check

References

• Perplexity getting started
• Get an API key
• Mirrored after crates/goose/src/providers/xai.rs.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9ad7f68986

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Mark PPLX_API_KEY as a valid configured credential

from_env explicitly accepts PPLX_API_KEY as a fallback, but metadata declares only PERPLEXITY_API_KEY as the required secret. Configuration status is computed from metadata via default_inventory_configured (crates/goose/src/providers/inventory/mod.rs), and exposed through provider_config_status (crates/goose/src/acp/server.rs), so users who set only PPLX_API_KEY will still be reported as unconfigured (and can be filtered out in configured-provider flows) even though runtime auth would succeed. Please add alias-aware configured checks (or include the alias in required-key semantics) so status matches actual provider behavior.

Useful? React with 👍 / 👎.