Live web index
Retrieve current web information for agents working with a changing environment.
Web search for reinforcement learning
For AI labs and frontier teams training agents to search. Retrieval shapes the evidence a model encounters during training and the context it receives at inference time.
A fixed web environment for RL.
Live webpages can change during a training run. Linkup pins the index so search uses a fixed web environment across rollouts.
The searchResults output returns source names, URLs, and content snippets. Deployment in your own infrastructure, index pinning, and the dedicated RL API are separate offerings to discuss with our team.
Explore search modes and API parametersThree ways to access the web
Pinned index
Keep the web index fixed across rollouts so changes to the index do not change the training environment.
Time machine
Query the web at a chosen date for RL evaluation and backtesting with historical context.
Why retrieval matters during RL
Retrieve pages with the answers
Give your agent evidence it can use to solve the task. Pages that contain the relevant facts give the model material to reason over at each step of a rollout.
Low cost at RL scale
The RL focused API provides dedicated resources at a lower price per query than our main API. For large training runs, search cost is part of the rollout budget from the start.
Build for training and inference
The retrieval environment shapes what your agent sees while learning and after deployment. Evaluate search at both stages so your training choices reflect the workflow you plan to ship.
Three ways to run your search infrastructure.
Your servers, your cloud account, or an API hosted by Linkup.
On premises
Your data center
Search index
Bring Your Own Cloud
Your cloud account
Search index
RL focused API
Hosted by Linkup
Dedicated RL resources