Introducing a new way to review your PRs.

Read the announcement

nousresearch / hermes-agent

The agent that grows with you

View on GitHub

38 documents · Updated Nov 20, 2026

overview

Knowledge Base updated: 2026-11-19

Hermes Agent knowledge base

Hermes Agent is a multi-surface AI agent product: a command-line control plane, long-running gateway, browser/desktop/TUI clients, and extensibility systems for tools, plugins, skills, and providers.

The gateway and agent runtime execute work; clients and platform adapters submit, observe, and deliver it. Durable state, sessions, profiles, messages, and coordination records are maintained by the shared persistence layer.

flowchart LR
  User[Users and platforms] --> CLI[CLI]
  User --> Web[Web dashboard]
  User --> Desktop[Desktop application]
  User --> TUI[TUI / ACP clients]
  CLI --> Gateway[Gateway service]
  Web --> Gateway
  Desktop --> Gateway
  TUI --> Gateway
  Gateway --> Runtime[Agent runtime]
  Runtime --> Providers[Model providers]
  Runtime --> Tools[Tools and guardrails]
  Gateway --> State[State and persistence]
  Gateway --> Platforms[Platform delivery]
  Runtime --> Plugins[Plugin platform]
  Runtime --> Skills[Skills system]
  Gateway --> Automation[Scheduled automation]

Glossary

  • Gateway — the long-running local service that accepts events, owns active sessions, and delivers agent output.
  • Profile — a named configuration and identity context used to select runtime settings, integrations, and models.
  • Turn — one bounded unit of agent work, from inbound request assembly through final output or failure.
  • Skill — reusable packaged instruction and support content that can be bundled or optionally installed for agent use.
  • ACP — Agent Client Protocol, the external protocol bridge for compatible agent clients.

Invariants

  • Session and message work must use durable shared state rather than client-local authority; see packages/state/** and docs/state-and-persistence.md.
  • Gateway clients interact through the generated RPC contract rather than independent ad hoc method definitions; see apps/…/gateway-contract.generated.ts and apps/…/gateway-contract.openrpc.json.
  • Agent execution routes tool use through approval, validation, and recovery controls; see packages/**/tools/** and docs/agent-tools-and-guardrails.md.
  • Plugin and skill extensions are discovered and validated before being exposed to runtime or product surfaces; see packages/plugins/**, skills/**, and their lifecycle pages.
  • Desktop packaging keeps the Electron shell and its provisioned payload aligned across build and relaunch paths; see apps/desktop/scripts/** and apps/desktop/electron/**.

Routing table