WISEnut

WISE iChat offers webhook style functional extension, so a customer can attach their own function to the chatbot. The reference deployments reach real systems. The Korea Expressway Corporation build works with the organization's internal messenger, and WISEnut names a high level understanding of legacy systems and crawling as part of how agents are grounded.

Document and vision ingestion pipelines handle varied formats through Document AI. The agents read from outside systems. The named reference work is answering from documents, summarizing, translating and searching, and WISEnut names no action where an agent writes to or changes state in a system of record. No connector catalog is published.

Sourceen.wisenut.com WISE iChat, WISE iRAG and AI Agent Sphere pagesread 2026-09-05

AI Agent Sphere is WISEnut's agent creation platform, the integrated body of its AI technology, and its stated aim is to create and combine various agents that work with people. WISE Agent Labs is the company's AI agent development and operations platform. Search Agents are built on WISE iRAG, and the reference deployments run real multi step work.

The Korea Expressway Corporation chatbot answers from laws, regulations and correspondence, summarizes and translates documents, and works with the organization's internal messenger. WISEnut says agents can be combined but does not say how. It describes no step model, execution runtime, branching, retry or handoff between participants.

Sourceen.wisenut.com AI Agent Sphere and WISE iRAG pages, wisenut.com newsroomread 2026-09-05

WISE iRAG retrieves enterprise internal data so the model answers from it. Hybrid search combines keyword retrieval for direct queries with vector retrieval for question type and long form natural language. Document AI handles varied formats and complex structures, with duplicate removal and optimal chunking. A reranker model is managed and served to improve accuracy, and search engine collection management covers both keyword and vector collections.

Embedding models are themselves trained against managed datasets to raise retrieval performance. The companion Vector Edition of Search Formula-1 uses the same combination, selected by query type and data characteristics. The result is an index that persists, scales past a context window and stays queryable. WISEnut calls it Korea's best. No refresh cadence or index scale is published, and in the named references the corpus is the customer's own document estate.

Sourceen.wisenut.com WISE iRAG and Vector Search pagesread 2026-09-05

Prompt management and answer simulation in RAGOps let operators author and test answers before release. WISEnut's stated aim of agents that work with people is a positioning line, not a shipped control. WISEnut names no step where a person reviews or approves agent output before it reaches someone. It also names no confidence threshold, handoff to staff, refusal policy or limit on what an agent may do.

Sourceen.wisenut.com AI Agent Sphere, WISE iRAG and WISE iChat pagesread 2026-09-05

Korean Cloud Service Security Certification, a government administered attestation for cloud services, is claimed on every page of WISEnut's site. WISEnut publishes no SSO, SAML, SCIM or OIDC, no RBAC, no roles and no granular permissions, so there is no published account of who inside the customer's organization can do what. RAGOps covers model, prompt and dataset administration, not user administration. WISEnut has no public product documentation site.

Sourceen.wisenut.com site footer and WISE iRAG pageread 2026-09-05

RAGOps is a GUI based operations system with integrated management across generative reasoning, model training, answer simulation, model administration and prompt management. An operator can see how the system is configured and what answers a configuration produces. All of this happens at design time.

WISEnut publishes no runtime trace, per session record, decision log, tool call history, retrieval provenance or audit surface, so a customer cannot reconstruct why a given production answer came out as it did or which documents it rested on. The named references are public bodies answering citizens from laws and regulations, where showing the basis of an answer matters.

Sourceen.wisenut.com WISE iRAG page, RAGOps sectionread 2026-09-05

WISE iChat handles multi turn conversation alongside machine learning intent analysis and its own five way conversation processing, so context carries across the turns of an exchange. The Korea Expressway reference adds a function called my own AI learning, which suggests adaptation per user, and WISEnut does not explain how it works.

WISEnut does not say whether anything is kept between separate conversations with the same person. It also does not say whether memory is scoped per user, per team or per workflow, how long it lives, where it lives, or whether an operator can inspect it or delete it for one person without touching the rest.

Sourceen.wisenut.com WISE iChat and WISE iRAG pagesread 2026-09-05

Customers can run WISEnut inside their own walls. A closed network, on premises design serves public institutions where security matters, and data does not move to an outside cloud but is processed only inside the institution. One form is an AI appliance that puts the WISE iRAG agent solution and the WISE LLOA domain model on a high performance NPU.

It works as an all in one search agent with no separate server build or model installation, and it runs once power and network are connected. WISEnut is also packaging its agent solutions to move away from individual bespoke builds, and its portfolio spans on premises AI agents through to the appliance. A hosted service is the other option, and it carries Korean Cloud Service Security Certification. Quantization and fine tuning change the model, not where it runs.

Sourcewisenut.com newsroom, en.wisenut.com WISE LLOA page and site footerread 2026-09-05

Customers assemble building blocks instead of adopting finished agents. The kit is WISE iRAG for retrieval, WISE LLOA as the domain model, Vector Search, WISE iChat for conversation, and WISE Agent Labs as the development and operations platform, from which Search Agents are built. WISEnut is packaging its agent solutions to move away from individual bespoke builds toward a faster supply structure.

The NPU appliance bundles iRAG and LLOA into an all in one unit that runs on power and network alone, and it is the nearest thing to a ready package. WISEnut publishes no set of prebuilt agents, templates or packs for a customer to select, and no gallery, catalog, named starter agent set or entitlement. The company applies 5,500 accumulated cases to produce optimized and customized agents, bringing its expertise to each build, not handing over assets.

Sourceen.wisenut.com AI Agent Sphere page, wisenut.com newsroomread 2026-09-05

Work reaches these agents when someone types a message. The channels are conversational. One is a web chat widget, which WISEnut runs on its own site, and in the named reference the other is the organization's internal messenger, with multi turn conversation carrying the exchange.

WISEnut names no scheduled run, event subscription, inbound webhook or queue, ticket intake or system state trigger, and no voice, email, SMS or social channel. The WISE iChat webhook runs outward as a functional extension and does not start work. WISEnut does not say which channels the WISE Answerny cloud service uses.

Sourceen.wisenut.com WISE iChat and WISE iRAG pagesread 2026-09-05

WISE iRAG connects flexibly to various LLMs, including WISEnut's own language model. The platform is not tied to one provider, and the customer makes the selection instead of a routing layer run by WISEnut. RAGOps includes model administration alongside prompt management and model training. Customers also get deep control over the chosen model.

WISE LLOA is a Korean specific model that supports additional learning, task learning and fine tuning on customer data. It offers lightweight variants made by quantization, which can be trained further, and domain specific customized reasoning models. Fine tuning and quantization shape one model, and the choice among models comes from the flexible LLM connection.

Sourceen.wisenut.com WISE iRAG and WISE LLOA pagesread 2026-09-05

WISE iChat offers webhook style functional extension, so a customer can hang their own function off the chatbot. RAGOps exposes prompt, dataset, collection and model administration through a GUI. WISEnut publishes no API or SDK for its own platform, and no developer portal, documentation subdomain, endpoint reference, client library or MCP server in either Korean or English. The webhook is the platform calling out to the customer, not the customer calling the platform.

Sourceen.wisenut.com WISE iChat and WISE iRAG pagesread 2026-09-05

RAGOps is a GUI based system for optimizing Search Agent development, with integrated management across generative reasoning, model training, answer simulation, model administration and prompt management. It covers dataset management for retrieval training, which improves embedding model performance against trained data, and search engine collection management for keyword and vector collections.

It also covers reranker model management and serving to improve search accuracy, prompt management for answer generation, dataset management for LLM training, and search and answer simulation. An operator changes an embedding model, a reranker or a prompt and simulates search and answers to see the effect. Optimization loops run both before and after deployment. WISEnut names no scored test set, pass rate, regression suite or published quality threshold, so RAGOps works as a tuning and simulation console without measured scoring.

Sourceen.wisenut.com WISE iRAG page, RAGOps sectionread 2026-09-05

The agents answer in chat and retrieve from an index. WISEnut names no control of a browser, a desktop or a remote computer. The reference deployments summarize documents, answer from regulations and work with a messenger, and none of that has an agent operating an interface. Crawling is among the techniques WISEnut names behind its understanding of legacy systems. Here crawling is index ingestion by a search engine, with WISEnut's own pipeline collecting content into a corpus it then queries. It works as a scheduled collector feeding an index, not as an agent working its way through a site.

Sourceen.wisenut.com AI Agent Sphere and WISE iRAG pagesread 2026-09-05