IP Library Granted Patent US 12,493,473
Granted Patent B1
US 12,493,473 · App. 19/234,232 · Granted Dec 9, 2025

Automated tool discovery and ingestion for artificial intelligence agents

Inventors: Roman Fedoruk (Cumming, GA); John Manton (Alpharetta, GA); Spencer Reagan (Marietta, GA); Gregory Roberts (Dunwoody, GA); Erich Stuntebeck (Johns Creek, GA)
Assignee: Airia LLC
G06F9/445G06F21/121
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,493,473
App. No.
19/234,232
Filed
Jun 10, 2025
Granted
Dec 9, 2025
Kind
B1
Art Unit
2433
USPC
726/26
Abstract

Systems and methods are described for tool discovery and ingestion for artificial intelligence (“AI”) agents. An AI platform can discover a first tool specification that includes an action and a description. The first tool specification is ingested to create a first tool object. The first tool object includes the action and an endpoint. Tool labels are determined from the specification and applied to the first tool object. A user interface displays the tool object, and it is added to an AI agent. The AI agent includes a manifest file that is used to execute the AI agent. This includes determining whether the AI agent is authorized to perform the action, and providing the AI agent with access to a tool credential, wherein the tool credential is sent to the endpoint.

Claims (53)

1 . A method for automatic tool ingestion for artificial intelligence (“AI”) agents, comprising:

discovering a first tool specification corresponding to a first tool, wherein the first tool specification comprises a tool action and a toot description, and wherein the first tool comprise service that an AI agent uses to perform specific actions;

ingesting the first tool specification to create a first tool object in an AI platform, wherein the first tool object comprises the tool action and a tool endpoint where the tool action can be invoked;

determining a plurality of tool labels that are associated with the first tool specification;

applying the plurality of tool labels to the first tool object;

causing a user interface (“UI”) of the AI platform to display the first tool object, including at least one of the plurality of tool labels;

generating, by the AI platform, a manifest file for the AI agent, comprising the first tool object, wherein the manifest file further comprises the tool action and the tool endpoint; and

deploying the AI agent for execution, wherein the AI agent is executed according to the manifest file, comprising:

determining that the AI agent is authorized to perform the tool action; and

causing a tool credential to be transmitted to the tool endpoint, wherein the tool credential authenticates the AI agent to initiate the tool action.

2 . The method of claim 1 , wherein determining the plurality of tool labels comprises determining a semantic meaning of the first tool specification.

3 . The method of claim 1 , wherein determining the plurality of tool labels is based on tool labels of other tool objects with semantically similar actions to the tool action of the first tool object.

4 . The method of claim 1 , wherein the plurality of tool labels comprises a name for the first tool object and a description of the tool action.

5 . The method of claim 1 , wherein the plurality of tool labels comprises a system prompt that is used during execution of the AI agent to determine when to perform the tool action.

6 . The method of claim 1 , wherein the tool object comprises the tool credential.

7 . The method of claim 1 , wherein the ingestion is performed by an ingestion agent, wherein the ingestion agent receives the tool specification and determines a semantic meaning of an application programming interface (“API”) description of the tool specification.

8 . The method of claim 1 , wherein the ingestion comprises ingesting code for local execution in performing the tool action.

9 . The method of claim 1 , wherein a first tool label of the plurality of the tool labels indicates an industry vertical.

10 . The method of claim 1 , wherein discovering the first tool specification comprises identifying the endpoint.

11 . The method of claim 1 , wherein discovering the first tool specification comprises scraping tool documentation from a website.

12 . The method of claim 1 , wherein discovering the first tool specification comprises making an application programming interface (“API”) call to request the first tool specification.

13 . The method of claim 1 , wherein discovering the first tool specification comprises subscribing to a service that provides tool specifications.

14 . The method of claim 1 , wherein the tool action is one of multiple application programming interface (“API”) calls included in an API definition of the first tool object.

15 . The method of claim 1 , wherein a system prompt is created as part of ingestion, and wherein, during execution, the AI agent determines whether to use the action based on the system prompt.

16 . The method of claim 1 , wherein the tool action is one of multiple tool actions viewable in the UI for the first tool object, and wherein the tool action is described by the at least one of the plurality of tool labels.

17 . The method of claim 1 , wherein the manifest file is automatically generated based on a UI selection to add the first tool object to the AI agent.

18 . The method of claim 1 , further comprising:

receiving user information from a first platform user;

automatically selecting a first template from a plurality of templates based on the user information; and

generating the AI agent based on the first template, including automatically adding the first tool object to the AI agent,

wherein the first tool object is part of an industry-specific toolset that applies to the first template.

19 . A non-transitory, computer-readable medium including instructions are executed by a processor and cause the processor to perform stages for automatic tool ingestion for artificial intelligence (“AI”) agents, the stages comprising:

discovering a first tool specification corresponding to a first tool, wherein the first tool specification comprises a tool action and a tool description, and wherein the first tool comprise service that an AI agent uses to perform specific actions;

ingesting the first tool specification to create a first tool object in an AI platform, wherein the first tool object comprises the tool action and a tool endpoint where the tool action can be invoked;

determining a plurality of tool labels that are associated with the first tool specification;

applying the plurality of tool labels to the first tool object;

causing a user interface (“UI”) of the AI platform to display the first tool object, including at least one of the plurality of tool labels;

generating, by the AI platform, a manifest file for the AI agent, comprising the first tool object, wherein the manifest file further comprises the tool action and the tool endpoint; and

deploying the AI agent for execution, wherein the AI agent is executed according to the manifest file, comprising:

determining that the AI agent is authorized to perform the tool action; and

causing a tool credential to be transmitted to the tool endpoint, wherein the tool credential authenticates the AI agent to initiate the tool action.

20 . A system for automatic tool ingestion for artificial intelligence (“AI”) agents, comprising:

a memory storage including a non-transitory, computer-readable medium comprising instructions; and

at least one hardware-based processor that executes the instructions to carry out stages comprising:

discovering a first tool specification corresponding to a first tool, wherein the first tool specification comprises a tool action and a tool description, and wherein the first tool comprise service that an AI agent uses to perform specific actions;

ingesting the first tool specification to create a first tool object in an AI platform, wherein the first tool object comprises the tool action and a tool endpoint where the tool action can be invoked;

determining a plurality of tool labels that are associated with the first tool specification;

applying the plurality of tool labels to the first tool object;

causing a user interface (“UI”) of the AI platform to display the first tool object, including at least one of the plurality of tool labels;

generating, by the AI platform, a manifest file for the AI agent, comprising the first tool object, wherein the manifest file further comprises the tool action and the tool endpoint; and

deploying the AI agent for execution, wherein the AI agent is executed according to the manifest file, comprising:

determining that the AI agent is authorized to perform the tool action; and

causing a tool credential to be transmitted to the tool endpoint, wherein the tool credential authenticates the AI agent to initiate the tool action.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jun 12, 2025
From: FEDORUK, ROMAN; MANTON, JOHN; REAGAN, SPENCER; ROBERTS, GREGORY; STUNTEBECK, ERICH
To: AIRIA LLC
Reel/Frame 071394/0270 →
Continuity (1)
Provisional Application 63658434 · Jun 10, 2024
References Cited (39)
US 9495133B1 · Righi et al. · 2016 [cited by applicant]
US 10009375B1 · Sites · 2018 [cited by examiner]
US 11616801B2 · Sjouwerman · 2023 [cited by examiner]
US 11748634B1 · Kulkarni et al. · 2023 [cited by applicant]
US 12021888B1 · Reed et al. · 2024 [cited by applicant]
US 12277245B1 · Morgan · 2025 [cited by examiner]
US 12282719B1 · Fedoruk et al. · 2025 [cited by applicant]
US 20110213712A1 · Hadar et al. · 2011 [cited by applicant]
US 20160063512A1 · Greenspan et al. · 2016 [cited by applicant]
US 20180152564A1 · Lang et al. · 2018 [cited by applicant]
US 20180173568A1 · El-Moussa et al. · 2018 [cited by applicant]
US 20180314971A1 · Chen et al. · 2018 [cited by applicant]
US 20190107968A1 · Wisnovsky et al. · 2019 [cited by applicant]
US 20190206390A1 · Rotem et al. · 2019 [cited by applicant]
US 20200257567A1 · Fontanari Filho et al. · 2020 [cited by applicant]
US 20200265509A1 · Kumar Addepalli et al. · 2020 [cited by applicant]
US 20220051112A1 · Wang et al. · 2022 [cited by applicant]
US 20220066905A1 · Lee et al. · 2022 [cited by applicant]
US 20220138004A1 · Nandakumar · 2022 [cited by applicant]
US 20220188691A1 · Katz et al. · 2022 [cited by applicant]
US 20230230001A1 · Draznin · 2023 [cited by applicant]
US 20230409654A1 · Ziv et al. · 2023 [cited by applicant]
US 20240184567A1 · Gao et al. · 2024 [cited by applicant]
US 20240202458A1 · Zha et al. · 2024 [cited by applicant]
US 20240281419A1 · Alfaras et al. · 2024 [cited by applicant]
US 20240296522A1 · Saito · 2024 [cited by applicant]
CN 115374515A · 2022 [cited by applicant]
WO 2019053488A1 · 2019 [cited by applicant]
Desfeux et al., Identification of a Series of Compatible Components using Artificial Intelligence, 2019, World Intellectual PropertyOrganization Patent Cooperation Treaty (PCT), pp. 1-34 (Year: 2019). [cited by applicant]
Ding et al, “Hybrid LLM: Cost-Efficient and Quality-Aware Query Routing”, arXiv preprint arXiv:2404.14618 (Apr. 2024). (Year: 2024). [cited by applicant]
Geronimo, “Evaluating LLMs with Semantic Similarity,” Mar. 2024 [retrieved on Nov. 30, 24], pp. 1-39, downloaded from :https://medium.com/@geronimo7/semscore-evaluating-Ilms-with-semantic-similarity-2abf5c2fadb9. (Year:… [cited by applicant]
Hazelwood et al., “Applied Machine Learning at Facebook: A Datacenter Infrastructure Perspective,” 2018 IEEE International Symposium on High Performance Computer Architecture (H PCA), Vienna, Austria, 2018, pp. 620-629,… [cited by applicant]
Lin et al., A Pipeline Design Method For Domestic Design(translation), 2022, Chinese Patent Office, pp. 1-8 (Year: 2022). [cited by applicant]
Lins et al, “Artificial Intelligence as a Service”, Bus Inf Syst Eng 63, 441-456 (2021 ). https://doi .org/10.1007 /s12599-021-00708-w (Year: 2021). [cited by applicant]
Marvin, “Prompt Engineering in Large Language Models,” Jan. 2024 [retrieved on Nov. 30, 2024], pp. 387-402, downloaded from :https://link.springer.com. (Year: 2024). [cited by applicant]
Minsuk, “LLM Comparator: Visual Analytics for Side-by-Side Evaluation of Large Language Models,” Feb. 16, 2024 [retrieved Mar. 19, 2025], pp. 1-7, downloaded from :https://arxiv.org/abs/2402.10524. (Year: 2024). [cited by applicant]
Niknazar et al. “Building a domain-specific guardrail model in production.” arXiv preprint arXiv:2408.01452 (Jul. 2024). (Year: 2024). [cited by applicant]
OnO “PipelineProfiler: A Visual Analytics Tool for the Exploration of AutoML Pipelines,” 2020 [retrieved on Nov. 30, 2024], pp. 390-400, downloaded from :https://ieeexplore.ieee.org. (Year: 2020). [cited by applicant]
Singh, An Autonomous Multi-Agent Framework using Quality of Service to prevent Service Level Agreement Violations inCloud Environment International Journal of Advanced Computer Science and Applications(IJACSA), 14 (3), … [cited by applicant]