Hermes-Agent UI

What Hermes-Agent UI does

Hermes-Agent UI is a TypeScript web front end with an API service for iterating on Hermes Agent locally. Reasoning traces arrive as structured frames (THOUGHT, TOOL_CALL, ARTIFACT, RESPONSE, STATUS, ERROR) with a monotonic sequence number, are batched before painting, and switch to windowed rendering at about 180 merged frames. The server persists replay data in SQLite and the client reconnects with resume_from_seq, using heartbeats and exponential backoff.

Access uses a JWT on the first WebSocket packet and a Bearer header over HTTP, with each session_id bound to a user_id and scope checks such as admin:stats:read. HTML artifacts render only inside a sandboxed iframe. The app shell has Chat, Insights, Skills and Orchestration areas, stores conversations in IndexedDB and handles interactive clarify prompts over one long-lived WebSocket. Docker Compose profiles separate a Demo setup from a Prod setup that runs non-root with an internal data plane. The README has a full Chinese mirror.

Key features

• Structured reasoning-trace frames with replay and resume_from_seq
• Windowed rendering for long traces
• JWT authentication with session ownership and scoped capabilities
• HTML artifacts rendered in a sandboxed iframe
• Chat, Insights, Skills and Orchestration areas with IndexedDB chat sessions
• Docker Compose Demo and Prod profiles

When to use it

• Watching a long Hermes reasoning trace without the browser slowing down
• Sharing a local Hermes UI between a few developers with per-user session ownership
• Previewing HTML artifacts the agent produces in an isolated frame

Who it is for: Individual developers and small teams who want a local web UI for Hermes Agent.

How it fits with Hermes Agent

Built for NousResearch/hermes-agent: it aligns with upstream tool contracts, including the interactive clarify flow.

Note: The README splits features into shipped and preview, so check its readiness table before relying on a given page.

FAQ