Case study
Hermes and OpenClaw: open agent frameworks, no lock-in
A business trigger enters the agent core, which is framework-agnostic and can run Hermes or OpenClaw with a swappable model, without rewriting the surrounding system. A human decision gate reviews and approves before any outward action. From the gate the flow fans out across multiple messaging channels at once, reaching people on WhatsApp, Slack, Microsoft Teams, and Telegram.
Context
Hermes and OpenClaw are both open-source, model-agnostic, multi-channel agent frameworks. Neither is a chatbot. Both can take actions, carry persistent memory, and connect to the channels a business actually uses. Nous Research built Hermes; OpenClaw was started by Peter Steinberger, formerly of PSPDFKit. We integrate and run them.
Most businesses asking about AI agents face a different problem: not what framework to pick, but how to get one running reliably, with real channels, human oversight, and no hard dependency on a single vendor. That is the gap we work in.
Hermes vs OpenClaw: the honest comparison
Hermes (Nous Research, MIT licence, Python) carries a self-improving skill loop that builds and refines capabilities from experience. Its heritage in the Nous Research model ecosystem means it integrates well with Hermes 3/4 models and similar open-weight models that use structured function-calling, though it works across many providers. Strong sandboxing options and research-grade Python extensibility make it the natural fit for teams who want deep control over how the agent learns and acts.
OpenClaw (TypeScript, open-source, local-first) takes a different stance. Its tagline is runs on any OS, any platform, and it means it. Channel coverage is very broad, including surfaces such as iMessage, Microsoft Teams, Matrix, and more alongside the usual messaging channels. A large and active community has grown around it. It suits teams who want the widest channel reach, a local-first agent that runs on their own devices, and a framework with strong community momentum.
Both are model-agnostic. Both support multi-channel dispatch. The choice comes down to stack, channel priorities, and how much Python versus TypeScript extensibility matters to the team.
What we do
We assess the use case, help choose the right framework, and integrate it as a live business agent connected to the channels the business actually uses. Whether that is WhatsApp for client communication, Slack or Teams for internal ops, or a combination, the agent meets people where they already are.
Human oversight at decision gates. The agent does not act without review at the points that matter. Gates are where a person approves before the system proceeds, so the work stays accountable.
Portability is designed in from the start. Switching framework or model later means re-testing and reconfiguration, not a full rewrite of the surrounding system. That design choice matters most when priorities shift or a better option appears.
Structured, validated outputs at each stage. What the agent produces can be inspected and checked rather than accepted on trust.
Outcome
Businesses get a live, multi-channel agent operation built on an open framework they understand, with human gates where they matter and no lock-in to a specific vendor or model.
For us, working across both Hermes and OpenClaw is the clearest proof of how we approach agent work: open frameworks, honest portability, people in control at the decision points. The same principles apply whether the agent is handling marketing, ops, or client communication.
“Nous Research built Hermes. Peter Steinberger's team built OpenClaw. We integrate and run whichever fits your use case, and we keep the system portable so you are not trapped by that choice.”
Integrated with: Hermes Agent (Nous Research, open-source) and OpenClaw (open-source), multi-channel dispatch, human-in-the-loop decision gates, model-agnostic orchestration.