Introduction to Runners
Runners are the engines that execute Tasks
Runners are the engines that run your task on our system. We have a range of standard runners built for different complexity of tasks: from simple & scalable web retrieval tasks to deep, robust research across many sources.
Any task can be run on any runner, making it easy to test and experiment on new tasks. When starting with a new task, we recommend running it across multiple runners to evaluate which best fits your needs.
Each Runner has different characteristics and our team will be happy to work with you to understand which runner is most applicable to your use cases. Available runners will be included in your onboarding guide. Please reach out to our team if you have any questions.
Runners, Processors, and Optimization
A Processor is a universal Runner that can be used to execute any Task.
If you are interested in creating a Runner that is specifically optimized for your Task, we have developed an Optimize feature for this purpose. It will generate sample outputs, collect your feedback, improve the outputs based on your feedback, and create a custom Runner (with a unique identifier) to use for future Runs.
Each time you execute a Task, you can choose any Runner or Processor. For example, you may use the Rhodium processor, or a custom Runner you created after Optimization.
Use Cases for Different Runners
Below is a non-exhaustive list of where you might use different Runners.
-
People Research (eg. find the CEO of
{company} and their background information): We have some Runners that are particularly good at finding people information. -
Investment Research (eg. identify the prior investments of
{company}): This would be considered a complex task and would be best run on a Runner that is optimized for deep research. -
Basic Company Information (eg. find the founding year of
{company}): This would be considered a simple task and would be best run on a Runner that is optimized for speed and scalability. -
News Tracking (eg. track the news about
{company}): This would be considered a Task that changes frequently and would be best run using Active Monitoring.
Available Runners
We have several available Runners. The following table provides a general guide to the typical response time and equivalent human research time for three of our Runners.
| Processor | p25 Latency | p90 Latency | Equivalent Human Research Time | Best For |
|---|---|---|---|---|
| neon | 10 sec | 84 sec | 5-15 min | Scalable and moderate complexity web research tasks, programmatic enrichment & classification |
| argon | 13 sec | 160 sec | 15-45 min | Deeper research with long-context (e.g., multiple web pages, PDFs, etc.), multi-hop, people research or complex web pages |
| rhodium | 15 sec | 180 sec | 1-3 hrs | Most complex use-cases requiring deep research with higher levels of quality, validation, etc. or larger output schemas |