We’ve been running Hermes Agent in production for months and the pattern is clear: the real value isn’t a one-shot coding prompt — it’s an agent that lives in your workspace, remembers your projects, and operates autonomously. Today we’re sharing the tools we built to make that repeatable.
The problem
Hermes Agent is open source, MIT-licensed, and runs on macOS, Windows, and Linux. It connects to Slack, Discord, Telegram, and more. It has persistent memory, scheduled automation, web browsing, delegation, and sandboxed execution. Out of the box it’s the most capable agent runtime we’ve used.
But getting it running as a durable, team-accessible service — especially one that touches production infrastructure — is work most teams don’t want to do twice. You need a container image that’s supply-chain verified. You need infrastructure. You need Slack app setup, credential management, persistent state. You need a way to update the agent without rebuilding everything.
That’s what bclaw.sh solves.
What is a bclaw?
bclaw stands for “business claw” — a long-running Hermes Agent instance deployed into your own AWS account, wired into your Slack workspace. Each bclaw is one agent with one identity: it has its own Slack app, its own ECS service, its own persistent storage.
You might create a @swe-pal that does code reviews and PR triage. Or a @reportclaw that posts scheduled status reports to configured channels. The point is that each claw is a purpose-built teammate, not a general-purpose chatbot.
How it works
Generate a repo. Deploy it. That’s the workflow.
npx @boldblackai/create-bclaw swe-pal
That command produces a swe-pal/ directory containing everything: CloudFormation templates for the AWS infrastructure, IAM policies scoped to exactly what the claw needs, Slack app setup instructions, and a full set of skills for setup, management, and teardown.
The generated repo is designed to be opened in harness — our sandboxed container runner — which verifies the Hermes Docker image signature and SLSA provenance on every launch before handing control to the agent. The claw runs on ECS with a persistent EBS volume for SQLite-backed state.
Once the infrastructure is deployed, you configure your inference provider (OpenRouter, Anthropic, or any provider Hermes supports) and the claw comes online in your Slack workspace.
The stack
The system has three layers, each open source:
Hermes Agent — the agent runtime. Persistent memory, multi-platform connectivity, scheduled tasks, delegation, web search, browser automation, vision, TTS, and a skill system that lets you encode reusable workflows. MIT licensed, maintained by Nous Research.
harness — the container sandbox. Wraps Hermes (and other agents) in a capability-dropped Docker container with cosign signature verification and SLSA provenance checks on every run. Skills, context files, and persistence are all handled automatically. Runs anywhere Docker runs.
create-bclaw — the deployment toolkit. An npx-distributed CLI that generates a complete, deployment-ready repository from a template. One command gives you the CloudFormation stack, IAM policies, ECS service definition, Slack integration, and all the operational tooling.
Why this matters
Most agent deployments today are either local-only (running on a developer’s machine) or fully managed (vendor-controlled cloud service). There’s a gap in between: teams that want the control of self-hosting without the operational tax of building the deployment themselves.
bclaw.sh fills that gap. You own the AWS account. You choose the inference provider. You control the skills and the agent’s personality. The deployment is opinionated enough to work out of the box but transparent enough to audit and modify — every template, every policy, every container image source is in the generated repo.
The harness image is signed and verified on every launch. IAM policies are least-privilege. State lives on an EBS volume you control. There’s no phone-home, no vendor lock-in, no data flowing somewhere you didn’t put it.
What’s next
We’re actively developing the bclaw toolkit and the harness container runtime. Current focus areas include additional deployment targets, more pre-built skill packs for common team workflows, and expanded provider support.
If you’re running Hermes Agent in a team setting — or want to — we’d like to hear about it. The project is on GitHub and you can reach us at hello@bold.black.
The code: github.com/boldblackai/create-bclaw · github.com/boldblackai/harness
The agent: hermes-agent.nousresearch.com