Skip to content
NagentNagent
Log inSign upHire your AI team
Atlas/Hermes Agent (Nous Research)
Vendor dossier / Autonomous watchlist

Hermes Agent (Nous Research)

Open, self-hostable general agent with tool use and long-running task execution; identity interpreted as Nous Research Hermes.

REVIEWED 23 SEPTEMBER 2026Marketing / Sales / CXNO HANDS-ON BENCHMARK

What it is

Official-source notesAutonomous watchlistwatchlist

Nous describes persistent memory, generated skills, scheduled tasks, messaging surfaces and isolated subagents. Hermes can run locally or with optional hosted services. [V1] [V2]

Documented means the statement is present in the cited material. Capability effectiveness, customer outcomes and security assurances have not been independently audited here.

Promoted use cases

  • Automate recurring research and briefings
  • Maintain task memory and reusable skills
  • Run extensible personal or custom agents

General agents matter because they change expectations about what an assistant can do. Their flexibility is also why an impressive personal demonstration does not establish readiness for shared enterprise work. This category is a watchlist: product identity, team access, permission boundaries and operational support need clearer evidence before treating an entry as a direct growth-platform substitute.

Pricing and buying model

Published commercial evidence

Free open-source core

MIT licence; model/hosting costs separate. Optional Nous Portal plans $20, $100 and $200/month.

Check current commercial source ↗

Budget for implementation, model and tool usage, data, human review and ongoing support. Credit units and outcome definitions differ between vendors. A missing numeric rate is marked unverified rather than replaced with an old third-party estimate.

Collaboration and governance

Sandboxing is described, but a self-hosted deployment still needs identity, secrets, policy and operational ownership. MIT licensing is not a hosted-service SLA.

The relevant unit of control is permission to interact with a connected application. Inspect who can propose, approve, execute, interrupt and audit that action. Shared seats, shared content and a shared live agent session are different capabilities; require a demonstration of the one your process needs.

Evaluate memory correction, permission revocation and release control for updated instructions. An improvement loop should preserve the original evidence, proposed change, test results and named approver. A safety or security badge alone cannot establish those workflow properties.

Review and customer evidence

No independent review sample was verified in this research pass. Vendor-hosted testimonials are selection-biased customer evidence, not an aggregate rating.

Before relying on a testimonial, confirm the exact product, version, package, workload and baseline. Ask a relevant customer about setup effort, failed cases, ongoing manual work and support after launch. This dossier does not convert customer logos or vendor-hosted awards into independent proof.

Focused feature documentation

Focused documentation check · Sep 2026

Open-source memory approval and scheduled delivery status

Hermes Agent documents persistent memory with an optional write-approval gate, allowing durable lessons to be staged before future sessions use them. Its scheduled-task documentation distinguishes successful execution from delivery failure, recording a failed send instead of calling the job complete. These are unusually concrete operational details for evaluating an autonomous agent, though they depend on the buyer’s own configuration and hosting.

Its security guide also discusses terminal and tool boundaries and secret handling; the plugin documentation states that third-party adapters do not automatically receive a blanket approval policy. A developer-led growth pilot should test a false memory correction, rejected write, scheduled outreach draft and failed delivery. MIT-licensed code is not zero operating cost: model, hosting, secret management, security review and ongoing maintenance remain with the operator.

Primary sources: Official persistent memory and approval guide ↗ · Official scheduled tasks ↗ · Official security guide ↗

Published documentation and public illustrations; account behaviour was not independently tested.

Visual source

Public Hermes Agent (Nous Research) source page captured 22 September 2026
Hermes Agent (Nous Research) · Public page captured 22 Sept 2026. Vendor presentation, not an authenticated product test. Open source ↗

Fit, limitations and proof requests

Editorial assessment

Hermes is an open implementation option and frontier watchlist entry. Multi-agent delegation is not by itself evidence of a governed multi-human workplace.

Priority question: Who reviews newly generated skills before they can execute, and how can a harmful memory be removed?

The evaluation owner should be an accountable pilot owner. Use useful completed work without unapproved side effects as the business target. Judge the solution in its own stack role: a cloud runtime, a specialist production tool and a managed AI team can be complementary purchases.

Use-case evaluation plan

A bounded research-and-artifact task

Choose a task using non-sensitive public inputs and require a sourced result with a clear completion condition. Include an ambiguous instruction and observe whether the agent asks an appropriate question or invents a direction. Keep the task limited enough that its output can be reviewed completely.

Measure: source accuracy, task completion and human correction effort.

Handoff between two humans

Where a team capability exists, ask a second authorized person to inspect progress and take ownership. Check whether account credentials, private context and pending actions remain correctly scoped. If the feature is undocumented, mark the result unknown rather than simulating multiplayer capability through shared passwords.

Measure: ownership clarity, access boundaries and continuity of task state.

Memory and unattended execution

Inspect how the agent stores useful facts and schedules repeat work. Correct an erroneous memory and revoke an integration before the next run. Test whether the change takes effect and whether a human can stop the process without losing the evidence needed to understand what happened.

Measure: memory correction, revocation behaviour and safe task interruption.

Detailed comparison

Read the detailed Nagent vs Hermes Agent (Nous Research) guide for operating models, shared context, governance, cost, evidence gaps and a staged pilot.

Nagent vs Hermes Agent (Nous Research)

Sources and evidence register

Nagent · Multiplayer AI for growth teamsResearch: 21–29 September 2026 · Public-source analysis, no hands-on benchmark · Sources & method