Which integrations are configured for a hosted Hermes Agent?
Hermes Agent is the runtime that invokes tools and model endpoints. Hermes Haven configures selected platform services around that runtime; HermesHaven operates the hosting and routing layer. An item on this page is therefore a hosted-service connection, not a claim that the HermesHaven platform itself is part of the open-source Hermes Agent project. [A-D6][C12]
The factual audit verified four useful connection categories: web search, web-content extraction, self-operated speech-to-text and model-provider routing. [C03][C23][C24]
Can an agent search and extract web content without customer API keys?
Yes. Hosted agents are configured to use a deployed SearXNG service for web search and Firecrawl for content extraction. The audited deployment supplies those service endpoints, so the customer does not need to provide keys for these two platform services. [C24]
| Connection | What is configured | Customer key required for this platform service? | Data-flow note | Evidence |
|---|---|---|---|---|
| SearXNG | Web-search backend | No | Search queries leave the agent and reach the platform search service; downstream engines may receive queries according to SearXNG configuration. The final downstream-engine detail was not established by the audit. | [C24] |
| Firecrawl | Web-content extraction backend | No | Requested URLs/content reach the platform extraction service. Its downstream fetch behaviour follows the requested site and current configuration. | [C24] |
| Speech-to-text | Proxy to operator-owned GPU hardware | No per-minute external API charge | Audio leaves the agent and is processed on operator-owned remote GPU hardware connected to the platform. Do not call it a European data-centre cluster. | [C23][C04] |
| Model providers | Platform and customer-configured inference routes | Depends on route | Premium or customer-configured routes may process prompts outside the EU. | [C03] |
How does speech-to-text work?
The audited speech-to-text path proxies audio to operator-owned GPU hardware. It does not depend on a per-minute external speech API, but the hardware is remote from the main platform cluster and connected by VPN. “Operator-owned GPU hardware” is supported; “European GPU cluster”, “data-centre GPU” and a specific city are not. [C23][C04][C02]
Do not translate “no external speech API” into “audio never leaves the agent”. The audio must reach the speech service to be processed. [C23]
Which model providers can be connected?
The audited configuration includes platform routes to external premium providers and supports customer-configured providers. The exact catalogue is time-sensitive and must be taken from a dated platform export before publication. [C03]
A provider connection changes the data boundary: prompts, tool outputs and other request content sent to that model can be processed under the provider's location, retention and policy terms. This is why the service must not claim that no data leaves the EU. [C03]
See routing and fallback details on Models and the security boundary on Security.
Are these integrations isolated from the internet?
No. Hosted agents have outbound internet connectivity, which is required for external tools and providers. Inbound access is routed through the platform HTTPS gateway, but that does not block outbound traffic. [C07][C22]
The per-agent overlay prevents direct agent-to-agent communication; it is not an allowlist of all external destinations. [C06][C07]
What should be checked before enabling a provider or tool?
- What content is sent: prompts, files, audio, URLs, tool output or metadata?
- Which service and downstream provider receive it?
- Where is it processed and retained?
- Which credential is required and who owns it?
- What happens when the service is unavailable?
- Does a plan, add-on or paid credit apply?
How do integrations relate to the managed platform?
The platform preconfigures selected shared services, but the agent remains the component that uses them. Review the full operational boundary on Platform, compare responsibility with a VPS on Compare self-hosting and VPS, and verify current commercial entitlements on Pricing.