Blog / data and privacy

Hermes Agent Privacy: Data, Credentials, and Controls

Hermes Agent privacy depends on four things: where the runtime runs, which model and tools receive the content, how credentials and persistent memory are stored, and who can access the deployment. Managed hosting removes infrastructure work but does not stop selected prompts from reaching a model provider. Self-hosting gives you direct control over the host but still depends on your provider choices, logs, backups, and operational discipline. Map those boundaries before giving an agent sensitive data or permission to act.

Updated Oct 20, 202610 min read

Deployment route

01
Classify the data
Separate public content, internal knowledge, personal data, credentials, and regulated records before connecting a workflow.
02
Map providers and tools
Record the messaging channel, model provider, search or browsing tools, and any storage or integration that receives the content.
03
Limit access
Use dedicated credentials, least-privilege channel roles, scoped tool access, and human approval for consequential actions.
04
Document retention and recovery
Know where logs, memories, files, and backups live, how long they are kept, and how they are removed or restored.

Four-step path

How do you evaluate Hermes Agent privacy?

Start with the data and the action, then trace every system that receives, stores, or can retrieve it.

Step 01
Classify the data
Separate public content, internal knowledge, personal data, credentials, and regulated records before connecting a workflow.
Step 02
Map providers and tools
Record the messaging channel, model provider, search or browsing tools, and any storage or integration that receives the content.
Step 03
Limit access
Use dedicated credentials, least-privilege channel roles, scoped tool access, and human approval for consequential actions.
Step 04
Document retention and recovery
Know where logs, memories, files, and backups live, how long they are kept, and how they are removed or restored.

What data does Hermes Agent process?

Hermes Agent can process account data, messages, prompts, files, memories, skills, configuration, integration credentials, logs, and usage events. The exact path depends on the channel, model, tools, and hosting mode you choose.

A private user interface does not automatically make the complete workflow private. If a message reaches an external model or tool, that provider is part of the data path.

Messaging content: messages and attachments sent to the connected bot.
Model content: prompts, context, tool output, and generated responses sent to the selected provider.
Persistent state: configuration, sessions, memories, skills, files, and schedules.
Credentials: channel tokens, model API keys, and connected-tool secrets.
Operational data: deployment status, errors, logs, and usage events.

Does Hermes Agent send data to model providers?

Yes, when the selected model is hosted by an external provider. The agent sends the context needed to generate a response, and the provider's retention and training terms then apply alongside the agent host's own policy.

A local model can keep inference inside infrastructure you control, but tool calls, browsing, email, storage, and remote integrations can still send data outside. Review the whole workflow rather than only the model logo.

For sensitive work, minimize the context, choose the narrowest provider path that meets the task, and avoid placing secrets inside ordinary chat messages.

How are channel and model credentials protected?

Use dedicated, least-privilege credentials and store them as secrets. This service encrypts channel credentials in its deployment storage, but any token can still become a liability if it is pasted into a message, exposed in logs, or reused across unrelated systems.

Create a dedicated bot or service identity instead of reusing a personal token.
Grant only the channel and tool permissions required by the first workflow.
Rotate a credential immediately if it appears in a message, screenshot, or log.
Separate test and production keys so an experiment cannot consume or expose production access.

What can Hermes Agent remember or log?

Hermes Agent can persist sessions, memories, skills, files, configuration, schedules, and operational logs depending on the deployment. Those records improve continuity but can also increase the impact of an account compromise or a request to delete data.

Decide what the agent is allowed to remember and which sources it may index. Keep secrets out of durable memory, review connected data sources, and document who can export, delete, or restore the workspace.

Is managed or self-hosted Hermes Agent more private?

Neither is automatically more private. Managed hosting centralizes platform operations with the provider, while self-hosting gives you direct control but also makes you responsible for patching, access, logs, backups, retention, and incident response.

Choose the model you can operate. A self-hosted deployment with an exposed dashboard, stale secrets, and untested backups is not safer than a managed service with clear boundaries. A managed service is not a substitute for reviewing model providers, channel permissions, or sensitive data policy.

For a regulated workflow, document the controller, processor, retention period, subprocessors, deletion path, and incident contact before launch.

Which primary sources support this page?

Product behavior is checked against the current deployment flow. These external links provide the upstream project or channel documentation used for setup details.

Direct answers

Frequently asked questions

Is Hermes Agent private by default?

No software can promise privacy without defining the deployment, providers, tools, retention, and access model. Review those boundaries for the exact workflow you plan to run.

Does self-hosting keep all data on my server?

Not necessarily. The runtime may stay on your server while prompts, tool calls, or files still go to external model and integration providers.

Can I delete Hermes Agent memories and account data?

The service policy describes account and deployment deletion requests subject to legal retention. Also remove data from connected model, channel, and tool providers according to their own policies.

Where is the legal privacy policy?

The Hermes Agent privacy policy is available at /privacy-policy. This guide explains the product workflow; the policy remains the authoritative legal document.

Key takeaways

Hermes Agent privacy depends on four things: where the runtime runs, which model and tools receive the content, how credentials and persistent memory are stored, and who can access the deployment. Managed hosting removes infrastructure work but does not stop selected prompts from reaching a model provider. Self-hosting gives you direct control over the host but still depends on your provider choices, logs, backups, and operational discipline. Map those boundaries before giving an agent sensitive data or permission to act.
Start with one model, one channel, and one controlled test conversation.
Use managed hosting when deployment speed matters more than operating the server yourself.

Continue the setup

Blog guide

Hermes Agent Privacy and Security: A Practical Guide

Learn how to protect Hermes Agent channel tokens, model keys, dashboards, memories, and team access in managed or self-hosted deployments.

Read next

Blog guide

Hermes Agent Credential Vault and Key Pools

Store model and tool credentials in the Hermes Agent credential vault, build key pools for rate limits, and rotate secrets without breaking the agent.

Read next

Use case