Hermes Agent adds LLM‑Wiki skill

Hermes Agent just got a new shortcut for one of the slowest jobs in artificial intelligence: turning a pile of raw material into something an agent can actually search. Nous Research says Hermes Agent now includes LLM-Wiki as a built-in skill, so the agent can generate research vaults and knowledge bases from web pages, code, and papers with a simple command. (x.com) That sounds small until you look at the usual workflow. Most teams still collect documents in one place, chunk them in another place, build retrieval indexes in a third place, and only then try to ask useful questions. A “knowledge base” here is not just a folder full of files. It is closer to turning a box of receipts into a labeled filing cabinet, where the agent can find the right page later instead of rereading the whole box every time. That is the bottleneck in many retrieval-augmented generation systems. Retrieval-augmented generation is the common pattern where a model first fetches outside material and then writes an answer grounded in that material, instead of relying only on what was inside the model at training time. The hard part is usually not fetching information from the web. The hard part is converting messy inputs into durable context that stays organized after the first query. Hermes Agent was already built around that larger idea. Nous describes it as a self-improving open-source agent with a built-in learning loop that can create skills from experience, improve them during use, persist knowledge, and search past conversations. (hermes-agent.nousresearch.com, github.com) Its skill system is designed to load instructions only when needed, which keeps token use down. In Hermes, skills are stored as on-demand knowledge documents and can be invoked as slash commands inside the command line interface or messaging platforms. (hermes-agent.nousresearch.com) That matters because LLM-Wiki fits naturally into the way Hermes already works. Instead of adding a separate external app, Nous appears to have turned the wiki-building workflow into something the agent can call like any other native skill. (x.com, hermes-agent.nousresearch.com) The phrase “research vault” points to a specific style of output. Rather than returning one disposable answer, the agent compiles an interlinked body of notes that can be searched, extended, and reused across later tasks. That is a different shape from classic question answering. Classic retrieval often acts like a librarian who runs to the shelf for every new question, while a wiki-style system acts more like a researcher building a permanent notebook as they go. There were already signs this direction was coming. In March 2026, a Hermes Agent feature proposal described a native “Code Wiki” skill that would transform a codebase into organized documentation with diagrams and source-linked explanations, instead of forcing users to inspect files one by one. (github.com) Another clue showed up in Hermes development on April 8, 2026. A feature request discussing per-skill model switching cited “llm-wiki” as an existing skill using Hermes skill configuration metadata, which suggests the wiki integration is not just a concept demo but part of the live skill architecture. (github.com) Nous demonstrated the new capability by having Hermes create a comprehensive knowledge base about Nous projects. The demo matters because it shows the skill working on a mixed research target with company projects, documentation, and likely code references, not just on a toy single-document example. (x.com) The timing also fits a broader push inside Hermes toward persistent memory and richer context handling. Hermes Agent v0.7.0, released on April 3, 2026, added pluggable memory providers and persistent session support, which makes the system better suited to storing and reusing structured knowledge beyond one chat window. (github.com) Put together, the update shortens a very specific loop: ingest source material, organize it, make it searchable, and use it in later agent work. For teams building research assistants, coding agents, or internal documentation systems, that loop is often where the real labor sits. The bigger shift is that wiki creation is moving from a side project to a first-class agent behavior. If an agent can read the web, inspect code, digest papers, and leave behind a reusable map of what it learned, the output stops being just an answer and starts looking more like infrastructure.