Skip to main content
Herm gives you persistent, self-learning agents through a REST API. Bring your system prompt, skills, tools, and MCP servers; Herm manages versioned agent definitions, user-scoped sessions and memory, persistent workspaces, credentials, and real-time event streams. The API is available at:

​
How it works

  1. Create an agent once with its model, system prompt, tools, MCP servers, and configured skills.
  2. Create sessions from that agent. Pass a stable subjectId to isolate each person, organization, workspace, or team.
  3. Send messages and stream events to render message deltas, tool calls, confirmations, status updates, and automatic session-title updates in your UI.
Infrastructure and deployment are managed for you. The API is centered on agents and sessions.

​
What you can do

Agents

Create, version, inspect, and archive reusable agent definitions.

Sessions

Start conversations with a stable subjectId identity scope.

Send Message

Send a message to a session and start or steer agent work.

Stream Events

Subscribe to deltas, tool calls, confirmations, status updates, and title changes.

​
Authentication

Send your workspace API key in the x-api-key header. All request bodies use JSON. Herm is a Prism product, so examples name the local environment variable PRISM_API_KEY.
Open Herm, create or select a workspace, then create a key under Settings → API keys. Keys are shown once; store them in your secret manager and never expose them in browser code. The machine-readable OpenAPI 3.1 specification describes the documented core REST surface. Product-specific integration endpoints may be released separately. Build your first agent with the Quickstart.
Last modified on July 25, 2026