Ethical Stack.

Hermes Agent

Open-source personal agent with terminal, browser, messaging, memory and task tools; the model and inference route are configurable.

Score calculation

(4 + 6 + 9 + 1 + 7) ÷ 5 = 5.4

5 of six dimensions rated. Unrated dimensions are excluded from the average.

Full scoring method ↗
Setup details

Hermes client and locally stored agent state. Hosted or local model route, messaging platform, plugins, and permissions determine data handling and action reach.

What fits your needs

Open-source harness installed on a user-controlled host; provider and enabled integrations determine the network route and access.

Task automation

Runs terminal, browser, messaging, memory and scheduled-task tools through one configurable agent.

See the evidence
Tradeoffs to expect

Configuration

You supply an inference provider and manage credentials, plugins, host security and integration permissions.

See the evidence
How to switch

Choose the model route

Use Hermes model setup to select a hosted provider or configure a local endpoint; local models can run on-device after model download.

See the evidence
Practical questions

Memory and approval

Built-in memory writes are ungated by default; `memory.write_approval: true` stages them for approval. Dangerous terminal commands prompt in interactive use and deny on timeout.

See the evidence

Inference route

Local models can keep inference on the machine after download; hosted providers receive requests under their terms. Model/provider and external memory are separately configured.

See the evidence

Energy practices

4/10↑

Hermes exposes local-versus-hosted inference choice and configurable token/context ceilings; no electricity mix or per-task energy is reported.

See the evidence

Its documented route selection and token/context limits define operational boundaries and give operators controls over input and output volume. They do not measure energy savings, identify a serving facility, or report electricity sourcing, so this is a practices score rather than a footprint claim.

1 + 1 reporting + 0 electricity + 2 controls = 4/10. This grades documented practices and transparency.

Actual token counts, local hardware consumption, and hosted provider/model/region electricity are not reported for a task.

Reporting1/3
Hermes documents both local-model execution and hosted provider routing, but gives no request-level energy or electricity disclosure.
Electricity sourcing0/3
The local-versus-provider choice does not identify the electricity mix or facility serving any selected model route.
Resource controls2/3
Documented absolute token compaction thresholds and context/file/output limits let an operator bound prompt and generated-token volume; these are controls, not measured energy reductions.

Water disclosure

Water reporting has not been assessed for this choice.

Training data

—Depends on the model↑

The product can route tasks through multiple models or providers, so no single training dataset applies to this agent entry.

See the evidence

The product can route tasks through multiple models or providers, so no single training dataset applies to this agent entry. Model version, provider, and any fine-tuning or post-training sources must be checked for the selected route; a client privacy setting doesn’t confirm training-material permission.

Exact model, version, provider, inherited base, and fine-tuning training data vary by route or are not disclosed in the reviewed product sources.

Your data & privacy

6/10↑

Hermes stores conversation history, memory, and skills locally; local inference can keep prompts on the device after model download, while cloud providers receive routed requests.

See the evidence

Hermes states it collects no telemetry and stores conversations, memory, and skills locally, and its managed local-model route can run without network access after downloads. Those controls apply to the client and local route; hosted provider, messaging integrations, plugins, and external memory change the boundary.

6/10 is a middle-band judgment: locally stored agent state and an available local inference route support user control, while hosted providers, gateways, integrations and backups remain outside that boundary.

The selected provider, external plugins, gateway host, messaging platform, and backups determine other copies and retention.

Openness

9/10↑

The NousResearch/hermes-agent repository is publicly available under MIT; connected model weights and services retain their own terms.

See the evidence

The official repository links to NousResearch/hermes-agent and publishes an MIT license for the agent code. Users can inspect and modify the harness, but model weights, paid gateways, plugins, and connected services are licensed separately and do not become open through the client license.

9/10 reflects the public MIT-licensed, inspectable harness; model weights, provider services and integrations keep separate terms and are not made open by the client license.

Bundled assets and each selected model/provider have separate licenses and disclosure limits.

Worker protections

1/10↑

Public project licensing does not disclose workforce conditions or contractor protections.

See the evidence

The reviewed official repository and project documentation identify the software and its license, but do not provide pay, working-condition, worker-voice, contractor, or independent-assurance evidence. This is a disclosure gap, not evidence of poor treatment.

1 + 0 + 0 + 0 = 1/10. Documented worker protections; missing disclosure is an evidence gap.

Nous Research workforce and outsourced annotation/moderation conditions were not established in the reviewed sources.

Pay & conditions0/3
Reviewed official project pages do not disclose a worker-conditions policy or workforce metrics.
Worker voice & remedy0/3
No worker-voice or remedy disclosures were identified in the reviewed official project sources.
Evidence & follow-through0/3
No independent workforce assurance is published in the reviewed project sources.
Workers rubric ↗

Military ties

7/10Not established↑

No military relationship or product-specific restrictions are established in the reviewed official materials.

See the evidence

This research found no official project disclosure establishing defense work or military safeguards for Hermes Agent. A configurable open-source harness can be paired with many model providers and deployments, so absence of a project disclosure doesn’t confirm how every user deploys it.

7 base + 0 safeguard adjustment = 7/10 (0 of 9 safeguard points; every 3 adds 1). No military relationship was established in the cited sources. This ranks above known defense ties without claiming a verified clean record.

Nous Research commercial relationships and downstream user deployments were not exhaustively audited.

Use restrictions0/4
The reviewed official repository and tool documentation do not state a binding military-use restriction.
Enforcement & oversight0/3
No military-specific enforcement or oversight outcomes are stated in the reviewed sources.
Relationship disclosure0/2
No military relationship disclosure was located in the reviewed official project sources.

The reviewed official project sources don’t show a defense contract, military deployment, or customer relationship.

Military rubric ↗

BDS targets

Not listed↑

Nous Research is not named in the reviewed BNC consumer guide or No Tech campaign.

See the evidence

The official BNC guide and technology campaign contain named company targets; Nous Research does not appear in the reviewed target listings. This limited check is not an endorsement and doesn’t confirm supplier or investor relationships.

This is limited to the two official BNC campaign sources; it does not cover every local campaign, supplier, investment, or future change.

More from Nous Research

All sources

Side by side