Does web search work with every model in the catalog?
Yes. Web search is implemented at the Backboard platform level, not as a tool call, so it works with any of the 17,000+ models you route through our API — including models that don't support function calling or tool schemas.
Is there a per-query cost for web search?
No. Web search is included in the Backboard platform at no additional charge. You pay for model usage and memory operations — not for search queries.
How does Backboard handle web results inside a limited context window?
Adaptive Context Management automatically budgets your context window. Web results are ranked by relevance and fit alongside your state, memory, RAG results, and system prompt — without overflowing the model's limit.
Do I need to wire a separate search provider?
No. You don't need to choose, integrate, or manage a third-party search API. Just set the web_search flag in your existing Backboard call and the platform handles querying, result selection, and formatting.
Can I combine web search with RAG and memory in the same call?
Yes. All Backboard features — routing, memory, RAG, web search, state management, and Adaptive Context — are available in a single API call. Backboard orchestrates them together and manages the context budget automatically.
What kinds of queries benefit most from web search?
Time-sensitive queries are the biggest win: recent model releases, benchmark results, news, pricing changes, and documentation updates that fall outside a model's training cutoff. It also helps with fact-checking model outputs against current public information.
Turn web search on with a simple flag
Add web_search: true to any existing Backboard message call. No new client, no separate endpoint, no configuration required — one flag and you're live.
Let Backboard handle querying, result ranking, and formatting
Backboard queries the web, selects the most relevant results, and formats them into context automatically. You never touch the search provider or parse raw results.
Feed the results into any model, even if it does not support tool calling
Web search is implemented at the platform level, not as a tool call. It works with all 17,000+ models — including those that don't support function calling or tool schemas.
HOW IT WORKS
How Backboard web search works
You enable web search as part of the same msg call you already use:
USE CASES
Web search patterns you can implement
Common patterns teams build once they have live web results available on every model call.
News‑aware assistants
Agents and assistants that answer questions about current events, market moves, or recent releases — grounded in live web results, not training data.
Hybrid knowledge bots
Combine RAG over internal docs with live web search for external coverage. One call returns answers grounded in both your proprietary content and the open web.
Research workflows
Multi-step agents that pull fresh source material on demand — benchmark data, pricing, papers, documentation — without leaving the Backboard API.
Safety and validation
Use web search to fact-check model-generated outputs against current public information before returning them to users. Reduce hallucination risk at the source.
COMPARISON
Why not just bolt on a search API?
Doing it yourself means choosing infra, stitching systems, and managing context budgets manually. Backboard handles all of it.
Doing it yourself usually means:
Choosing and integrating a third‑party search provider
Designing result formatting for each model
Handling rate limits, errors, and different relevance formats
Backboard gives you:
Web search integrated into the same unified, stateful API
Free usage as part of the platform — no per‑query fees or separate search bill.
Automatic context budgeting alongside routing, memory, and RAG
You toggle a flag instead of building another subsystem. Backboard fits web results into the context window automatically, alongside everything else.
PLATFORM
Included in Backboard, not a separate product
Web search is a first‑class, free feature of Backboard.