A prompt is the basic unit of work in the Mantium pipeline - this is a specific text processing configuration. In the UI, prompts are accessible in the AI Manager menu.
With Mantium, it is possible to chain together multiple prompts so that the output from one prompt feeds into the input of the next. This is referred to as an intelet.
The term intelet itself was coined by an AI when we asked it, "What do you call a group of AI's?"
The provider is the company or organization that builds and maintains the Large Language Model you want to use.
Our current supported providers:
All prompts and intelets configured in Mantium belong to an organization. If you are a single user, the organization and your user identity are synonymous. If you are working in a team, there may be several people who are all members of your organization and have access to the same set of prompts and intelets.
It is also possible to be a member of more than one organization. If so, you need to select which organization you are working with in order to see or interact with that organization's prompts and intelets.
Mantium offers integration with providers to create prompts using your individual provider API key.
Click here to learn how to obtain keys.
Please note: Mantium does not supply provider API keys.
Notification integration currently supports email and Slack notification to send prompt alerts.
Click here to learn how to create a Slack integration.
A security policy is a set of security rules that can be applied while the prompt is processing. When any rule in the policy is tripped, Mantium will take the actions that are specified with the policy.
You can have multiple policies configured for a single prompt. They will be processed in order. Within a policy, all of the rules are applied in parallel.
Security policies are accessible in the Security menu.
A security rule is a specific check in the text processing pipeline. For example, you may wish to set a limit on the number of tokens that any prompt input is allowed, or to set a rate limit for how many requests can be made to a prompt in a specified timespan.
Security rules are applied via a security policy, which may implement several rules at the same time.
Actions can be configured inside of security policies. If a prompt violates its set security policy, the selected action will take place.
Log warning - all policy violations are automatically logged and can be viewed in the Monitoring menu.
Halt processing - immediately stops processing of the prompt.
Interrupt processing - prompt processing is temporarily interrupted until it is restarted using the Human In the Loop (HITL) feature.
Mantium security policies have the option of pausing prompt processing until a human can make a decision about the violation. When configured, the human will be notified. They can then choose to 'Accept', 'Reject', 'Modify', or 'Reprocess' the prompt. See the Human In the Loop section of this documentation for more information about how that works.
Updated over 1 year ago