Abstract: A method for identifying and clustering worker agents for processing requests includes receiving, by a core node, from a user agent, a user request. The core node updates, for each of the plurality of worker agents, an availability status, thereby producing a plurality of availability statuses. The core node computes, for each of the plurality of worker agents, a value of a drift metric. The core node clusters the plurality of worker agents to produce a plurality of clusters of worker agents, wherein each of the plurality of clusters contains worker agents that have similar semantic capabilities. Based at least on the user request, the plurality of availability statuses, and the plurality of clusters, the core node identifies a subset of the plurality of worker agents that are both available to process the user request and that are suitable for processing the user request.

Abstract: A method for verifying a user agent includes transmitting, by a core node, to a user agent, a predefined set of user requests. For each user request in the predefined set of user requests, the core node receives an embedding generated based on the user request, thereby receiving a plurality of embeddings based on the predefined set of user requests. The core node determines whether the plurality of generated embeddings satisfy a similarity criterion. The core node transmits, to the user agent, a task to generate a query plan and receives a query plan generated by the user agent for processing a user request identified in the task. The core node determines whether the query plan satisfies a query plan criterion, thereby producing query plan adequacy output. The core node determines whether to approve the user agent based on the similarity output and the query plan adequacy output.

Abstract: A method for verifying a user agent includes transmitting, by a core node, to a user agent, a predefined set of user requests. For each user request in the predefined set of user requests, the core node receives an embedding generated based on the user request, thereby receiving a plurality of embeddings based on the predefined set of user requests. The core node determines whether the plurality of generated embeddings satisfy a similarity criterion. The core node transmits, to the user agent, a task to generate a query plan and receives a query plan generated by the user agent for processing a user request identified in the task. The core node determines whether the query plan satisfies a query plan criterion, thereby producing query plan adequacy output. The core node determines whether to approve the user agent based on the similarity output and the query plan adequacy output.

Abstract: A method includes receiving a user request from a user. A user agent identifies a current worker agent capable of handling a current part of the user request and of satisfying the set of requirements of the current part of the user request. The current worker agent processes the current part of the user request to generate and send a response to the user agent, which parses the response to extract relevant context information. The user agent enables the extracted relevant context information to be appended to subsequent user requests. The user agent selects, based on an additional part of the user request, an additional worker agent to process the additional part. An augmented user request is constructed based on the additional part of the user request and the extracted relevant context information.

Abstract: A method includes recording user-agent interaction data associated with a plurality of interactions between a user and a user agent, the user-agent interaction data including a plurality of requests provided by the user to the user agent, a plurality of responses provided by the user agent, and feedback provided by the user in response to at least some of the plurality of responses. The user-agent interaction data is summarized to produce summarized user-agent interaction data, which is stored in a scalable database optimized for quick retrieval. A user request is received and parsed to identify an intent of the user request. A similarity search of the user request is performed against the summarized user-agent interaction data, thereby identifying a matching subset of the summarized user-agent interaction data. The matching subset of the summarized user-agent interaction data is concatenated with the user request to produce an enhanced context.

Abstract: A method for ranking a plurality of worker agents based on a user request includes receiving a user request, the user request including a description of a task. The method incudes identifying a plurality of tasks to be performed to process the user request. The method includes, for each of a plurality of worker agents, computing a corresponding value for each of a plurality of metrics, thereby computing a plurality of metric values for each of the plurality of worker agents. The method includes generating, for each of the plurality of worker agents, a corresponding ranking, based on the metrics computed for the plurality of worker agents above, thereby generating a plurality of worker agent rankings, each of which corresponds to a distinct worker agent in the plurality of worker agents.

Abstract: A method for verifying a worker agent includes receiving, by a core node, from a worker agent, a capability description describing a plurality of tasks and, for each of the plurality of tasks, (i) at least one parameter of the task and (ii) an outcome expected to be produced by performing the task. The method includes generating, based on the capability description, a plurality of request-output pairs, each representing a particular request and a corresponding baseline output expected to be produced upon processing the request. The core node receives, from the worker agent, a plurality of outputs, each of the plurality of outputs generated by the worker agent and corresponding to one of the plurality of request-output pairs. The core node compares the plurality of baseline outputs to the plurality of actual outputs to produce comparison output and determines whether to approve the worker agent based on the comparison output.

Abstract: A method for identifying and clustering worker agents for processing requests includes receiving, by a core node, from a user agent, a user request. The core node updates, for each of the plurality of worker agents, an availability status, thereby producing a plurality of availability statuses. The core node computes, for each of the plurality of worker agents, a value of a drift metric. The core node clusters the plurality of worker agents to produce a plurality of clusters of worker agents, wherein each of the plurality of clusters contains worker agents that have similar semantic capabilities. Based at least on the user request, the plurality of availability statuses, and the plurality of clusters, the core node identifies a subset of the plurality of worker agents that are both available to process the user request and that are suitable for processing the user request.

Abstract: A method for assigning a plurality of worker agents to perform a plurality of sub-tasks includes generating, by a user agent, an embedding of a user request using a first pre-trained large language model. A core node in communication with the user agent identifies and ranks a plurality of candidate worker agents based on the embedding of the user request. A second LLM decomposes the user request into a plurality of sub-tasks. For each of the plurality of sub-tasks, the user agent identifies a subset of the plurality of candidate worker agents to perform the sub-task and identifies a best worker agent from the subset. For each of the plurality of sub-tasks, the user agent assigns the sub-task to, and receives a corresponding result from, the best worker agent; the user agent provides the corresponding result to the user.

Abstract: A method for verifying a worker agent includes receiving, by a core node, from a worker agent, a capability description describing a plurality of tasks and, for each of the plurality of tasks, (i) at least one parameter of the task and (ii) an outcome expected to be produced by performing the task. The method includes generating, based on the capability description, a plurality of request-output pairs, each representing a particular request and a corresponding baseline output expected to be produced upon processing the request. The core node receives, from the worker agent, a plurality of outputs, each of the plurality of outputs generated by the worker agent and corresponding to one of the plurality of request-output pairs. The core node compares the plurality of baseline outputs to the plurality of actual outputs to produce comparison output and determines whether to approve the worker agent based on the comparison output.