An assistant authorized to act<\/strong> : call up tools, link steps together, The database that stores the vectors of document fragments and A set of job instruction sheets, loaded on demand by an agent. A tool that an agent can call: web search, file reading, query The amount of text a template can process at once. It includes the The splitting of a document into fragments of uniform size, with a Transforming text into a vector \u2014 a list of numbers representing its The organizational unit of WivenLLM: a set of documents, settings, and An injected document in full<\/strong> in the context of each message, without An independent conversation within a workspace. The threads A sequence of steps defined in advance and always executed in the same A piece of document resulting from the cutting process. This is the unit that the search A statement confidently produced by a model, but false or fabricated. The engine that writes the answers. WivenLLM is not one: it orchestrates the An open standard describing how an AI application discovers and uses A short fact remembered about you and reintroduced into conversations. The assistant uses your documents And<\/strong> his general knowledge.<\/p>\n The assistant replies uniquely<\/strong> based on your documents, and refuses The mode that enables named accounts, roles, shared spaces, web widget and API.<\/p>\n The permanent instructions given to the model before each message: role, tone, The central mechanism: finding the relevant passages in your documents, the A mechanism that automatically directs each conversation to the model Your personal knowledge base, which spans all areas of The minimum degree of proximity that a fragment must reach to be retained by The document excerpts displayed below an answer indicate what it is about. An agent's instruction executed automatically according to a schedule. The degree of unpredictability of the responses. Low for factual responses, high for...
\nto produce a concrete result. Triggered by @agent<\/code> in a conversation.
\n\u2192 Agents Verification Status<\/a><\/p>\nVector base<\/h3>\n
\nfinds the closest matches to a question.
\n\u2192 Vector base and embeddings<\/a><\/p>\nSkills library<\/h3>\n
\nThis should be distinguished from skills, which involve executing code.
\n\u2192 Skills library<\/a><\/p>\nSkill<\/h3>\n
\nSQL, calendar\u2026 Activated by an administrator, then by workspace.
\n\u2192 SKILLS<\/a><\/p>\nContext (window of)<\/h3>\n
\nprompt system, pinned documents, retrieved passages, history and
\nThe question. When it is saturated, the oldest elements are discarded.<\/p>\nCutting<\/h3>\n
\nrecovery, to allow for research.<\/p>\nEmbedding<\/h3>\n
\nMeaning. Two texts that are close in meaning produce similar vectors.<\/p>\nWorkspace<\/h3>\n
\nConversations. Two spaces share nothing.
\n\u2192 Workspaces<\/a><\/p>\nPinned (document)<\/h3>\n
\nUse the search function. This should be reserved for short and consistently relevant documents.<\/p>\nThread<\/h3>\n
\nThey share documents and space settings, but not their history.<\/p>\nAgent flow<\/h3>\n
\norder. Unlike an agent, who decides his own path.
\n\u2192 Agent workflows<\/a><\/p>\nFragment (chunk)<\/h3>\n
\nfinds and which is provided to the model.<\/p>\nHallucination<\/h3>\n
\nThe Query mode and source citations serve to contain and
\ndetect.<\/p>\nLLM (Large Language Model)<\/h3>\n
\nthat you choose.<\/p>\nMCP (Model Context Protocol)<\/h3>\n
\nexternal tools. \u2192 MCP Servers<\/a><\/p>\nMemory<\/h3>\n
\n\u2192 Memoirs<\/a><\/p>\nChat Mode<\/h3>\n
Query Mode<\/h3>\n
\nexplicitly when they do not contain the answer.<\/p>\nMulti-user mode<\/h3>\n
Prompt system<\/h3>\n
\nFormat, prohibited. The application's most effective lever.
\n\u2192 Conversations<\/a><\/p>\nRAG (Research-Augmented Generation)<\/h3>\n
\nprovide the model, then generate the response from them.<\/p>\nModel router<\/h3>\n
\nmore suitable, according to rules.
\n\u2192 Model router<\/a><\/p>\nSecond Brain<\/h3>\n
\nwork. \u2192 Second Brain<\/a><\/p>\nSimilarity threshold<\/h3>\n
\nResearch. Too high, it excludes what is useful; too low, it lets in noise.<\/p>\nSources<\/h3>\n
\nrest. They allow for verification.<\/p>\nPlanned task<\/h3>\n
\n\u2192 Planned tasks<\/a><\/p>\nTemperature<\/h3>\n
\ncreativity.<\/p>\nWeb widget<\/h3>\n