fix(config): respect configured model

README.md

@@ -56,9 +56,10 @@ Once installed, the agent automatically calls `perplexity_search` whenever it ne
 | `query` | string | ✅ | The search query |
 | `recency` | string | — | Filter by age: `hour` · `day` · `week` · `month` · `year` |
 | `limit` | number | — | Max sources to include (1–50) |
-| `model` | string | — | Model preference, e.g. `pplx_pro_upgraded`, `pplx_pro`, `experimental`, `gpt54`, `gpt54_thinking`, `claude46sonnet`, `claude46sonnetthinking`, `gemini31pro_high`, `nv_nemotron_3_super`, `pplx_reasoning`, `pplx_alpha` |
 | `incognito` | boolean | — | Whether to hide the search from Perplexity history; defaults to `true` |
 
+Model selection is configured globally with `/perplexity-config` or `PI_PERPLEXITY_MODEL`; it is not exposed as a tool parameter, so agent-generated tool calls cannot accidentally override your configured model.
+
 ### Output format
 
 The tool returns structured text the agent can reason over:
@@ -98,6 +99,13 @@ bun test           # Run tests
 bunx tsc --noEmit  # Type check
 ```
 
+Optional live model-selection E2E test (requires cached auth from `/perplexity-login`):
+
+```bash
+PI_PERPLEXITY_E2E=1 bun test test/e2e-models.test.ts
+PI_PERPLEXITY_E2E=1 PI_PERPLEXITY_E2E_MODELS=pplx_pro_upgraded,gpt54 bun test test/e2e-models.test.ts
+```
+
 ## License
 
 MIT — see [LICENSE](LICENSE) for details.