Can Hermes Agent tailor your CV to a job?

What did Nous Research actually launch on October 7?

Nous Research confirmed a $90 million Series B at a $1.5 billion valuation on October 7, 2026, and used the announcement to launch Hermes for Businesses — an enterprise tier of its open-source Hermes Agent that lets companies deploy customised agents for multi-step workflows while keeping their data private. Robot Ventures led the round, with NVIDIA, Microsoft's M12, Samsung, Y Combinator, Union Square Ventures and Menlo Ventures among the backers. That brings total funding to $158 million for a lab founded in 2023, following a $50 million Series A in April 2025. SSO and compliant access controls are promised for enterprise customers, with a mobile app on the roadmap.

The agent itself stays free and MIT-licensed, which is why it spread so fast. By the company's own count, Hermes Agent has been cloned more than 24 million times and now drives roughly 2.5% of global AI token usage. It runs locally or in the cloud, handles search, coding and image understanding out of the box, and — the part that matters for job seekers — keeps a persistent memory across sessions and writes its own reusable skills from experience. It isn't a chat tab you open and close. It's a process that remembers what you told it last Tuesday.

Every general-purpose agent that launches gets pointed at the same first task by ordinary users: fix my CV. That instinct is sound. Reading a job ad, extracting what it really asks for, and rewriting your experience to speak that language is genuinely good work for a model with memory and file-handling skills. But CV tailoring has a second half that agents quietly skip — proving the document you exported survives automated screening and a recruiter's thirty-second scan. The rest of this piece covers what Hermes Agent does well on the first half, and the three checks you still have to run yourself.

On October 7, 2026, Nous Research confirmed a $90 million Series B at a $1.5 billion valuation and launched Hermes for Businesses.

Can Hermes Agent tailor your CV to a job description?

Yes — Hermes Agent handles the drafting side of CV tailoring competently, and its persistent memory makes it better suited to a long job search than a one-off chat session. Point it at the text of a job ad and ask it to list every hard requirement, tool, certification and responsibility, ranked by how often each appears. Then have it compare that list against your existing CV and flag what's missing. Because the agent keeps a working memory of your career, you only tell it about the migration you led in 2024 once. On application forty, it still knows.

The workflow that gets real mileage looks like this. Keep one master file of everything you've ever done — projects, numbers, tools, team sizes, awkward failures you learned from. Let the agent read it, then ask for a tailored version per role: a reordered skills section, five rewritten bullets that use the ad's own vocabulary, and a short summary that names the job title you're applying for. Its document skills can export to Word or PDF, and scheduled tasks mean it can check saved searches overnight and have a shortlist waiting for you at breakfast.

Here's the contrarian bit: the writing was never the hard part. Any decent model produces plausible bullets. The hard part is knowing whether your claims hold up and whether the file you attached is readable. An agent writing in markdown or a text buffer sees clean prose; it has no idea that your template pushed your job titles into a text box the parser ignores. If you want the same keyword-extraction logic applied to the document a recruiter actually receives, upload the file for a free CV analysis and compare what comes back.

Yes, for drafting and keyword extraction — but it can't check the file it produces.

Which three checks will Hermes Agent never run for you?

Three checks sit outside what any general-purpose agent can do: parsing verification on your exported file, a realistic account of why you'd be rejected, and a score against the market you're competing in. Each one requires something the agent doesn't have — a parser reading your actual PDF bytes, a recruiter's pattern library, and comparison data from other candidates. Hermes Agent can tell you your bullets are stronger than they were. It cannot tell you that a hiring manager will bin the CV in eight seconds because your most relevant role is buried on page two under a hobbies block.

The parsing gap is the costliest. Two-column layouts, icons standing in for contact details, skills inside shapes, text in headers and footers, and tables used for alignment all degrade extraction. The agent writes content; your template destroys it on export, and nothing in the agent's feedback loop catches that. This is why a single-column CV a parser can read end to end still beats a beautiful one. Visual layout analysis — looking at how the page is actually built, not just what it says — is a separate job from writing, and it needs a different tool.

The second and third gaps are judgement calls dressed up as feedback. Ask an agent whether your CV is good and it will say yes, because agreeable answers are what it's trained to produce. A proper rejection read-out names the blind spots: an unexplained eighteen-month gap, a title that doesn't match your scope, a stack list with nothing from the last three years. A market score goes further and estimates how many interviews ten applications should yield, plus a salary band for the role. HRLens puts the score out of 100, five category scores and an ATS compatibility rating in the free tier; the rejection reasons, blind spots, hiring probability and salary estimate sit in the $3.99 Full Analysis.

Parsing verification, rejection reasons, and a score against the market.

How do you build an agent workflow that actually ends in interviews?

Build it as four stages — gather, extract, draft, verify — and never let the agent own the last one. Start by writing a single evidence file: every role with dates, every tool you've touched, every number you can defend in an interview. Feed that to the agent once and let its memory hold it. Stage two is extraction: paste the job ad's full text rather than a link, because LinkedIn and Indeed block automated fetching and you'll get an empty page or a login wall instead of the requirements. Twenty seconds of copy-paste saves you a hallucinated job description.

Stage three is the draft, and this is where specific prompts beat vague ones. Ask for the ad's requirements ranked by frequency, then for a gap list, then for bullets rewritten one at a time with the metric kept intact. If you want a tested set of instructions to adapt, the prompt sequence for mining achievements and tailoring to a job ad transfers cleanly to Hermes Agent, since it's model-agnostic and runs 300-plus models through swappable endpoints. Different models have different strengths here — the model-by-model breakdown for resume tasks is worth a look before you pick a backend.

Stage four is verification, and it has to happen on the exported file. Download the PDF, upload it, and read the ATS compatibility score and layout analysis before you attach it to anything. If the structure is beyond saving — a 2019 template, three columns, skills in a sidebar — rebuilding beats patching, and a chat-based CV builder with six ATS-friendly templates will get you there faster than fighting Word. Then log the application. An agent with memory and scheduled tasks is genuinely excellent at tracking what you sent where, and chasing the follow-up you'd otherwise forget.

Gather, extract, draft, verify — and the agent doesn't own verification.

Will a recruiter spot that an agent wrote your CV?

Often, yes — and the tells are stylistic, not technical. Recruiters reading three hundred applications a week develop a fast ear for machine rhythm: every bullet the same length, every one opening with a power verb, "spearheaded" and "orchestrated" appearing in a junior analyst's history, and metrics that sound impressive but round suspiciously to 30%, 40%, 50%. Agent-written CVs also tend toward uniform vocabulary across unrelated jobs, because the same model drafted all of them. A human career has texture. A model smooths it out, and smoothness is the giveaway.

None of this makes using an agent a problem. Using one badly is. The fix is editing rather than generating: take the agent's draft, then put back the specifics only you know — the name of the legacy system you migrated, the headcount you inherited, the quarter the project slipped and why. Break the bullet rhythm deliberately. Keep one or two short, blunt lines. Replace every rounded percentage with the real figure, even if it's an untidy 17%. If you're unsure how your draft reads from the other side, a check for the phrases that give AI-written drafts away is a useful sanity pass.

The deeper point is about where the agent's judgement runs out. Hermes Agent learns from your interactions and refines its own skills, which means it gets better at the task you've defined — but it's optimising against your feedback, not against a hiring manager's. Nobody in that loop has ever rejected a candidate. A screening-aware check closes the gap, because it's measuring the thing that decides your outcome: whether the file parses, whether the content answers the ad, and whether a recruiter sees the relevant work in the first eight seconds.

Often yes — and the tells are stylistic, not technical.

Frequently asked questions