Three self-hosted agent platforms with different centers: a chat assistant, an agent runtime, and a team workspace.
OpenClaw, Hermes Agent and Buzz all run on infrastructure you control and connect to models you choose. They solve different problems. This page compares what each one is for, how it installs, how agents and tools connect, what the security model assumes, and what a business has to run after launch.
Quick answer
OpenClaw is a personal assistant that lives in your chat apps through one Gateway. Hermes Agent is a provider-flexible agent runtime built around the CLI, toolsets and profiles. Buzz, from Block, is a self-hosted team workspace where people and approved agents share channels on a signed event log. We set up each one for $3,000 plus $400 a month maintenance.
Choosing between OpenClaw, Hermes Agent and Buzz?
Free. Three short fields, and we reply by email within one business day.
At a glance
| Dimension | OpenClaw | Hermes Agent | Buzz | Verdict |
|---|---|---|---|---|
| What it is | A personal AI assistant whose Gateway connects models, sessions, tools and chat channels | A general agent runtime run through a CLI, messaging Gateway, API server or editor protocol | An Apache 2.0 team workspace by Block where people and agents share channels on a Nostr relay | Start from the job: an assistant in chat, a controlled agent process, or a shared team room. |
| Install path | Official installer for macOS, Linux and Windows; Node 26 recommended on the npm path | Desktop installer on macOS and Windows, plus CLI install scripts and a Docker guide | Docker Compose bundle for one VPS, a Helm chart, or a one-click Railway relay | All three need an always-on host, a non-root account, backups and a rollback plan for business use. |
| Where people use it | Core chat channels such as WhatsApp, Telegram, Slack and Discord, plus channel plugins | 20+ messaging platforms in its docs, with the CLI central to setup and operations | Buzz desktop app and CLI; mobile clients are still being built | Confirm the one channel you need and how approvals work there before choosing. |
| Agents and models | Connects hosted and local model providers you configure | Provider-flexible, including custom OpenAI-compatible endpoints | Agents such as Goose, Codex and Claude Code connect through the ACP harness with their own keys | None bundles a private model. Every provider you connect is a data processor to review. |
| Tools and MCP | Tool policy, skills, plugins and MCP servers; sandboxing is off by default | Toolsets per platform or session, skills, MCP with per-server tool filters | Agents use the Buzz CLI; the ACP harness can pass one optional MCP server to the agent | Pick the platform where you can express the exact permission policy the job needs. |
| Security model | One trust boundary per Gateway; separate Gateways for untrusted parties | One process is one trust boundary; never run a production Gateway as root | Signed logins and channel membership as the only access gate; TLS must be added in front | None is a multi-tenant boundary. Separate installs for people who should not see each other's data. |
| Maturity | Moves quickly; check the docs and pin versions before each update | Moves quickly; keep risky unattended modes on their default deny settings | Pre-1.0; only the latest main branch gets active security support | Pin versions and test updates before rolling them out on any of the three. |
| Our setup | $3,000 setup plus $400/month maintenance | $3,000 setup plus $400/month maintenance | $3,000 setup plus $400/month maintenance, after a deployment review | The fee is the same. The right choice depends on the work, the channel and who will run it. |
When a managed platform is enough, and when your workflow needs owned code and infrastructure.
Both add operational capacity, but their judgment, supervision, and cost structures are different.
MCP commonly uses existing APIs; it adds a standard interface for compatible AI applications.
Deployment control matters, but no workflow platform is HIPAA-compliant by configuration alone.
Whether to pay a managed service to run your AI agents or hire someone to build and run them on staff.
Whether to hire a firm or an independent freelancer to build your automations and AI agents.
What each one charges for, where your data runs, and when owned code is worth the upfront cost.
Free. Three fields. We reply within one business day.
Tell us who will use the agent and where, and we'll recommend one and quote the setup.
What happens next
“Shariq built our Open Dental to GoHighLevel sync and our AI routing system for new patient inquiries. Both projects were delivered fast, worked right away, and he was responsive whenever we needed adjustments. He understands dental practice workflows and doesn't overcomplicate things.”
Talk to Shariq
Pick whichever is easiest.
We reply within one business day. You talk to the person who builds it.
Prefer writing? Get a written plan, no call
