Deploy AI agents withOne API call.
Backed byCombinatorDeploy a fully harnessed AI agent for every customer. Bring your system prompt, skills, tools, and connectors — get infrastructure, orchestration, memory, self learning, and more out of the box.
Problem
Building a good AI agent is hard
A production agent is more than a model in a loop. Before customers trust it with real work, it needs every layer below — and on a bare SDK, each one is a project you build and maintain yourself, none of which makes your product different.
Tool integration
Hand-rolled MCP clients, tool schemas, and per-customer credential injection — with allowlists to scope as your tool surface grows.
Skills
Domain playbooks crammed into the system prompt burn context on every turn. On-demand loading needs a skill format the model actually follows.
Automations
Cron parsing, job queues, retry backoff, and run state that survives a worker dying mid-job — for every schedule a customer creates.
Memory
Context compaction inside a session, persistence across them, and per-tenant scoping — thread checkpoints alone can't do it.
Self learning
Turning corrections into future behavior takes idle-time consolidation and feedback loops you'd design from scratch.
Infrastructure
Container-per-customer isolation, persistent volumes, SSE fan-out with replay, secret vaulting — plumbing measured in months.


Product
Create an agent in one API call
Stop harness-engineering. Create a versioned agent with your system prompt, skills, tools, and connectors, then start an isolated, subject-scoped session for each user — with memory, automations, self-learning, and a persistent filesystem built in.
Reusable versioned configuration
Read docsDurable threads and runs
Read docsTokens and tool calls over SSE
Read docsLearns every user
Read docsScheduled background runs
Read docsMarkdown playbooks on demand
Read docsWeb, files, and shell built in
Read docsYour tools, open protocol
Read docsA persistent workspace
Read docsCredentials, never in the sandbox
Read docsHuman approval mid-run
Read docsIsolated per agent and subject
Read docsImproves between sessions
Read docsRuns survive crashes
Read docsHerm vs. building it yourself
The harness is the moat you shouldn't build
Every capability below is mandatory before customers trust an agent with real work — and none of it differentiates your product. Building it yourself means months of harness engineering and a maintenance treadmill. Herm provides the managed runtime when you create an agent, with sessions and customer-scoped resources exposed through the API.
- Agent + infrastructure in one API call
- Durable execution — runs survive crashes and deploys
- Session management and compaction
- Observational memory across sessions
- Per-customer container isolation
- Credential management for connectors
- Steering and human-in-the-loop approvals
- Streaming over SSE
- Built-in tools: web search, shell, file operations
- Versioned Markdown skills
- Automations and scheduled runs
- Self-learning and dreaming
- Persistent goals (Ralph Wiggum loop)
- New harness features without a rewrite
Herm— versioned agent configuration and managed runtime, built on the Hermes open-source harness.
Book a demoFrequently Asked Questions
Everything you need to know about Herm — the API, the pricing, and what you can build.
Contact SupportWhat is Herm?
Herm is an API for deploying user-facing AI agents built on the open-source Hermes harness. Create a reusable agent with your system prompt, skills, tools, and connectors, then create a subject-scoped session for each user and chat with it over SSE. We run the containers, the harness, and everything in between.
What is Hermes?
Hermes is the open-source personal agent from Nous Research with 185k+ GitHub stars. It ships with session management, observational memory, self-learning, skills, scheduled automations, isolated subagents, real sandboxing, a persistent filesystem, and full browser and web control. It's the same harness powering products like Higgsfield Supercomputer.
Herm lets you create Hermes instances programmatically — the agent and the infrastructure in a single API call — so you focus on the engineering that's unique to your product.
Do I need to set up any servers or infrastructure?
No. One POST /v1/agents call creates your agent, and Herm runs its sessions in isolated containers with a persistent filesystem and credential management. There's no harness to build, no Docker to wrangle, and no EC2 instance to babysit.
How do I integrate it with my product?
Bring your system prompt, MCP servers, configured Markdown skills, and tools to POST /v1/agents. The API returns an agent ID; create a session with POST /v1/sessions, then wire its SSE events endpoint into your chat UI. Your users get memory, automations, self-learning, and image & video input without you building any of it.
How is this different from LangChain Managed Deep Agents or Claude Managed Agents?
All three give you an agent and infrastructure in one API call. Herm additionally exposes automations, persistent goals (the Ralph Wiggum loop), steering, dreaming, and image & video input — with no provider lock-in, so you choose the model.
What can my users' agents do?
Anything Hermes can do: browse the web, search, manage files, run your tools, follow your skills, and schedule recurring work. A user can say “every week, look at our top-performing video from last week and make five variations” and the agent handles it in the background.
What happens to my users' files between sessions?
Every agent has a persistent filesystem. Files, memory, skills, and automations survive across sessions, so the agent picks up exactly where it left off — even days later.
Is my users' data private?
Yes. Every user's agent runs in its own isolated container. Secrets are managed through credential references rather than raw tokens, and Herm does not train on your data.
Can I self-host?
Yes. Book a call to discuss private and hybrid deployment options.
Ship your agent today
Bring a system prompt, skills, tools, and connectors — get a dedicated AI agent for every customer, with an endpoint to chat over SSE. Herm runs the infra.