> ## Documentation Index > Fetch the complete documentation index at: https://exa.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. > ## Agent Instructions > The Exa API is served at https://api.exa.ai. Authenticate with `Authorization: Bearer $EXA_API_KEY` (or `x-api-key: $EXA_API_KEY`); create keys at https://dashboard.exa.ai/api-keys?utm_source=docs&utm_medium=docs&utm_campaign=markdown_instructions. Every way an agent can get access to Exa is described at https://exa.ai/auth.md. > Prefer the official SDKs, `exa-py` (`pip install exa-py`) and `exa-js` (`npm install exa-js`); both read `EXA_API_KEY` from the environment. > Tool-using agents can call Exa without writing code through the hosted MCP server at https://mcp.exa.ai/mcp, or install the Exa agent skill with `npx skills add exa-labs/agent-skills` (skill file: https://exa.ai/docs/skill.md). > The OpenAPI specs at https://exa.ai/docs/exa-spec.yaml and https://exa.ai/docs/team-management-spec.yaml are the source of truth for request and response schemas. # OpenRouter > Ground any OpenRouter model with Exa web search through the openrouter:web_search server tool. Exa is the search engine behind [OpenRouter](https://openrouter.ai)'s web search. OpenRouter gives you one API for hundreds of models, and Exa gives those models live web access: any model without native search grounds through Exa by default, and any model can be pointed at Exa explicitly. No Exa API key is needed. OpenRouter runs the searches server-side and bills them to your OpenRouter credits. ## Use the web search server tool Add `openrouter:web_search` to your `tools` array and the model decides when to search, what to search for, and whether to search again within the same request. [Server tools](https://openrouter.ai/docs/guides/features/server-tools/web-search) are in beta on OpenRouter, and they replace the deprecated `web` plugin and `:online` model variants; see OpenRouter's [migration guide](https://openrouter.ai/docs/guides/features/server-tools/web-search#migrating-from-the-web-search-plugin) if you use either. ```javascript JavaScript theme={null} const response = await fetch("https://openrouter.ai/api/v1/chat/completions", { method: "POST", headers: { Authorization: "Bearer ", "Content-Type": "application/json", }, body: JSON.stringify({ model: "openai/gpt-5.2", messages: [ { role: "user", content: "What were the major AI announcements this week?" }, ], tools: [{ type: "openrouter:web_search" }], }), }); const data = await response.json(); console.log(data.choices[0].message.content); ``` ```python Python theme={null} import requests response = requests.post( "https://openrouter.ai/api/v1/chat/completions", headers={ "Authorization": "Bearer ", "Content-Type": "application/json", }, json={ "model": "openai/gpt-5.2", "messages": [ {"role": "user", "content": "What were the major AI announcements this week?"} ], "tools": [{"type": "openrouter:web_search"}], }, ) print(response.json()["choices"][0]["message"]["content"]) ``` ```bash cURL theme={null} curl https://openrouter.ai/api/v1/chat/completions \ -H "Authorization: Bearer " \ -H "Content-Type: application/json" \ -d '{ "model": "openai/gpt-5.2", "messages": [ { "role": "user", "content": "What were the major AI announcements this week?" } ], "tools": [{ "type": "openrouter:web_search" }] }' ``` With the default `engine: "auto"`, OpenRouter uses the provider's native search where a model has one and Exa everywhere else. Set `engine: "exa"` to keep one search behavior across every model: ```json theme={null} { "type": "openrouter:web_search", "parameters": { "engine": "exa", "mode": "auto", "max_results": 5, "max_total_results": 20, "allowed_domains": ["arxiv.org"], "excluded_domains": ["reddit.com"] } } ``` | Parameter | Use it to | | - | - | | `mode` | Trade latency for depth: `instant`, `fast`, `auto` (default), `deep-lite`, `deep`, or `deep-reasoning`. The modes map to Exa's [search types](https://exa.ai/docs/search/quickstart.md). | | `max_results` | Cap results per search call (default 5) | | `max_uses` | Cap how many times the model may search in one request | | `max_total_results` | Cap cumulative results across all searches in one request | | `max_characters` | Set an exact per-result character budget for highlights | | `search_context_size` | Use a preset budget instead: `low`, `medium`, or `high` | | `allowed_domains`, `excluded_domains` | Filter result domains. Exa supports both filters in the same request. | `instant`, `fast`, and `deep-reasoning` are [legacy search types](https://exa.ai/docs/search/quickstart.md#legacy-search-types). ## How results come back OpenRouter requests [Exa highlights](https://exa.ai/docs/search/highlights.md) for each result rather than full page text: extractive excerpts sized adaptively, typically 2,000 to 4,000 characters per result, unless you set `max_characters` or `search_context_size`. The model reads the excerpts, and API callers receive them in standardized `url_citation` annotations on the response message. Within one result, `[...]` markers separate excerpts drawn from different parts of the page. ## Pricing Exa searches bill to your OpenRouter credits, in addition to the model's token costs for reading the results. The `instant` mode costs \$0.004 per search, `fast` and `auto` cost \$0.007, `deep-lite` and `deep` cost \$0.012, and `deep-reasoning` costs \$0.015. Each search includes up to 10 results, and additional results cost \$0.001 each. See [OpenRouter's web search docs](https://openrouter.ai/docs/guides/features/server-tools/web-search) for current rates. The response's `usage` object reports how many searches the model ran in `server_tool_use.web_search_requests`. ## Resources > [card] [Server tool docs](https://openrouter.ai/docs/guides/features/server-tools/web-search) > > Full configuration reference for `openrouter:web_search`. > [card] [Customer story](https://exa.ai/customers/openrouter) > > How OpenRouter gives hundreds of models web search with Exa.