@IBuzovskyi: HERMES AGENT NOW FACT-CHECKS ITS OWN RESEARCH. EVERY CLAIM BACKED BY A VERIFIABLE SOURCE. EVERY QUOTE MATCHED AGAINST T…

X AI KOLs Timeline Products

Summary

Hermes Agent v0.20.0 introduces grounded citations that link every research claim to a specific source and passage, plus a fact-checking mode that categorizes claims as confirmed, unverified, or contradicted.

HERMES AGENT NOW FACT-CHECKS ITS OWN RESEARCH. EVERY CLAIM BACKED BY A VERIFIABLE SOURCE. EVERY QUOTE MATCHED AGAINST THE ACTUAL PAGE TEXT. SHIPPED IN v0.20.0. before grounded citations: you ask Hermes to research a topic. it returns a summary. sounds confident. you have no way to know if the claims are real without opening every source yourself. the agent could hallucinate a quote, invent a statistic, or misattribute a finding. you'd never know unless you checked manually. after grounded citations: hermes skills install official/research/grounded-citations every claim the agent makes now links to the specific page and passage it came from. quotes are matched against actual page text. if the quote doesn't exist on the page, it doesn't make it into the output. WHAT IT LOOKS LIKE: you ask: "research the latest Hermes release" old output: "Hermes v0.20.0 has 3,650 commits and introduces conversational voice." (no source. trust me bro.) new output: "Hermes v0.20.0 includes 3,650 commits [1] and introduces conversational voice with barge-in [1]." [1] github. com/NousResearch/hermes-agent/releases/tag/v2026.8.3 every number traceable. every claim sourceable. click the link. find the exact passage. if it's not there, the agent didn't include it. FACT-CHECKING MODE: hand it any document, article, or claim. "fact-check this article about GPT-5.6 pricing. verify every number against primary sources." the agent reads the document. checks each claim against the web. returns three categories: CONFIRMED: claim matches primary source. source linked. passage quoted. UNVERIFIED: no matching source found. could be true. could be fabricated. flagged for your manual review. CONTRADICTED: source says something different. both versions shown. you decide. WHY THIS MATTERS: research is only useful if it's trustworthy. an agent that writes confident summaries with no source trail is a liability. grounded citations turn Hermes research from "the agent thinks this is true" to "here is where it found this." the difference between an intern who says "I read somewhere that..." and one who says "page 12, paragraph 3, here's the link." USE CASES: competitive research: "research competitor X pricing. cite every number from their actual pricing page." content creation: "draft a post about Opus 5 benchmarks. every benchmark number must link to the official announcement." due diligence: "fact-check this pitch deck. verify every market size claim and every competitor comparison." article verification: "here is my draft article. check every technical claim against official documentation." HOW TO ENABLE: hermes skills install official/research/grounded-citations or tell your agent: "install the grounded-citations skill" works with any model. works in CLI, Desktop, and Telegram. once installed, every research task includes source attribution by default. no extra prompting needed.
Original Article
View Cached Full Text

Cached at: 08/10/26, 03:33 PM

HERMES AGENT NOW FACT-CHECKS ITS OWN RESEARCH. EVERY CLAIM BACKED BY A VERIFIABLE SOURCE. EVERY QUOTE MATCHED AGAINST THE ACTUAL PAGE TEXT. SHIPPED IN v0.20.0.

before grounded citations:

you ask Hermes to research a topic. it returns a summary. sounds confident. you have no way to know if the claims are real without opening every source yourself.

the agent could hallucinate a quote, invent a statistic, or misattribute a finding. you’d never know unless you checked manually.

after grounded citations:

hermes skills install official/research/grounded-citations

every claim the agent makes now links to the specific page and passage it came from. quotes are matched against actual page text. if the quote doesn’t exist on the page, it doesn’t make it into the output.

WHAT IT LOOKS LIKE:

you ask: “research the latest Hermes release”

old output: “Hermes v0.20.0 has 3,650 commits and introduces conversational voice.” (no source. trust me bro.)

new output: “Hermes v0.20.0 includes 3,650 commits [1] and introduces conversational voice with barge-in [1].” [1] github. com/NousResearch/hermes-agent/releases/tag/v2026.8.3

every number traceable. every claim sourceable. click the link. find the exact passage. if it’s not there, the agent didn’t include it.

FACT-CHECKING MODE:

hand it any document, article, or claim.

“fact-check this article about GPT-5.6 pricing. verify every number against primary sources.”

the agent reads the document. checks each claim against the web. returns three categories:

CONFIRMED: claim matches primary source. source linked. passage quoted.

UNVERIFIED: no matching source found. could be true. could be fabricated. flagged for your manual review.

CONTRADICTED: source says something different. both versions shown. you decide.

WHY THIS MATTERS:

research is only useful if it’s trustworthy. an agent that writes confident summaries with no source trail is a liability.

grounded citations turn Hermes research from “the agent thinks this is true” to “here is where it found this.”

the difference between an intern who says “I read somewhere that…” and one who says “page 12, paragraph 3, here’s the link.”

USE CASES:

competitive research: “research competitor X pricing. cite every number from their actual pricing page.”

content creation: “draft a post about Opus 5 benchmarks. every benchmark number must link to the official announcement.”

due diligence: “fact-check this pitch deck. verify every market size claim and every competitor comparison.”

article verification: “here is my draft article. check every technical claim against official documentation.”

HOW TO ENABLE:

hermes skills install official/research/grounded-citations

or tell your agent: “install the grounded-citations skill”

works with any model. works in CLI, Desktop, and Telegram.

once installed, every research task includes source attribution by default. no extra prompting needed.

Similar Articles

NousResearch/hermes-agent

GitHub Trending (daily)

Hermes Agent is an open-source, self-improving AI agent framework by Nous Research featuring a closed learning loop, cross-platform deployment, and compatibility with hundreds of LLMs. It provides a terminal interface, persistent memory, automated scheduling, and research-ready tooling for scaling AI workflows.