Automatically inject live web search results into every prompt before the model generates a response.
Not yet available. The Web Search plugin is not yet available.
The Web Search plugin is a request-phase plugin that analyses your prompt, decides whether a web search would improve the answer, runs the search, and injects the results into the context — all before the model generates a single token. Unlike the Web Search server tool, the model never makes an explicit tool call; the plugin acts transparently.
The plugin inspects the last user message, scores it for search relevance, and conditionally injects results. Factual or current-events questions trigger a search; creative or purely reasoning tasks do not.
| Option | Type | Default | Description |
|---|---|---|---|
threshold | number | 0.6 | Minimum relevance score to trigger a search (0–1) |
max_results | integer | 4 | Number of search snippets to inject |
region | string | "dz" | ISO 3166-1 alpha-2 region for result ranking |
inject_position | string | "before_user" | Where to inject: "before_user" or "system_append" |
| Web Search Plugin | Web Search Server Tool | |
|---|---|---|
| Model awareness | Transparent | Model sees tool calls |
| Trigger | Plugin heuristic | Model decision |
| Suitable for | Always-on grounding | Selective, model-driven retrieval |
| Prompt overhead | Fixed, predictable | Variable |
For most always-on use cases, the plugin is simpler. Use the server tool when you want the model to decide when a search is necessary.
A flat search fee applies each time the plugin triggers a search, plus the token cost of injected results. No fee is charged on turns where the plugin's relevance check scores below threshold.