Users and groups
Waystone comes with built-in user management and tracking for your LLM application.
When you make a model request through Waystone, you can optionally provide a unique identifier for the end-user triggering that request to allow Waystone to handle per-user usage tracking. This allows you to:
- Set quotes on users, groups, or a combination (e.g. no single user in group X can take more than 10% of monthly credits)
- Monitor model usage to track down service abuse
- Apply special rules on important users (additional model output checks; model routing etc.)
Entities
Waystone allows you to define:
user
(individuals)group
(e.g. orgs or teams)- a
user
and agroup
Once a request is sent with both a user
and a group
,
Waystone creates a link between the entities: “user is part of group”;
such requests are attached to both the user and the group,
and is subject to rules and conditions
on both.
Once a user is attached to a group, however, any request with just the user provided will only be applied to that user.
Metadata
You can attach arbitrary metadata to each user and group,
which is then used to apply rulesets
.
Useful metadata for your application may include:
- region
- payment plan / tier of product
- categories of note, such as “risky” or “vip”
Tracking usage
To attach a request to an entity,
provide the unique ID for the user or group in your application
when making the request.
The Waystone API is a superset of the OpenAI API,
so the metadata
field is used to also provide user identification in user requests:
In this example, the request is attached to a user, which is simulatanously given some additional metadata.
Refer to the API for more information.
Taking Action on Users
Waystone’s concept of rulesets
allows you to automatically take actions on users or groups
by their metadata.
Rulesets
Learn how to group users and take action on user requests by their metadata.
Exploring usage
Functionality around exploring entity usage is under development with Waystone beta users. This includes:
- OTEL on model inputs and outputs
- Tracking usage (tokens; requests; USD cost)
- Usage analysis (spikes, abuse detection, etc.)
To gain access to usage tracking, please get in touch.