Updates without gambling production
Hermes moves quickly. We test stable releases on our own deployment first, then update yours in an agreed window. The goal is simple: you do not become the person debugging an upstream change on Monday morning.
SELF-HOSTED AI AGENTS
Self-hosted AI agents for your company, implemented with Hermes
We install Hermes Agent on infrastructure you control, connect it to the tools your team already uses, set the permissions and cost limits, and stay responsible for updates and operations.
The install is the easy part.
Costs you can explain
We choose which jobs need an expensive model, set budgets per agent, and report usage. Local models are an option where the task and hardware make sense, not a promise we make before testing the real workflow.
Permissions before autonomy
Connections start read-only where possible. Risky actions can require approval. Agent execution can be isolated from the rest of the server, and credentials stay outside prompts and procedure files.
Skills that do not become a junk drawer
A skill is a reusable procedure the agent keeps. We write the first ones from your actual process, then review what the agent adds or changes so the library stays useful instead of accumulating instructions nobody trusts.
One agent system, split by responsibility.
Where your team talks to it
Slack, Microsoft Teams, WhatsApp, Telegram, email or another supported channel. We connect the places your team already uses instead of asking people to adopt a second inbox.
Separate agents for separate jobs
Sales, operations and finance can run as separate Hermes profiles, each with its own memory, skills, credentials and chat history. Access boundaries are configured around the job, not around one giant company bot.
The model mix
Routine work can run on a local or lower-cost model. Steps that need stronger judgment can use a cloud model under a defined budget. We test the split on the client's task before recommending it.
Company memory and procedures
Recurring work becomes written procedures the agent can reuse. Memory and skills live in accounts and infrastructure the client controls, with backups and review rather than a black box nobody can inspect.
Connections to your tools
CRM, project management, files, email, ERP or internal APIs. We start with the minimum access the workflow needs and widen it only when there is a reason to.
The operating rules
Approval gates, spend limits, logs, backup and recovery are part of the deployment. The agent is allowed to do specific jobs under specific conditions, not simply given broad access and told to be careful.
When Hermes is the wrong tool
If the work is a deterministic flowchart with stable rules, n8n or a conventional integration is often cheaper. If your existing OpenClaw deployment is stable and already does the job, we would not migrate it just because Hermes is newer. And if nobody owns the workflow or can name the metric it should improve, we would fix that before installing an agent.
Our filter is simple: use Hermes when the work needs judgment, tool use, memory and procedures that change over time. Use simpler automation when it does not.
One workflow. Thirty days to prove it.
01.
Map the task
Week 1
We sit with the person who does the work today and write down the inputs, sources, forbidden actions, expected output, approval points and success metric. That procedure is the first deliverable.
02.
Install and fence it
Week 2
Hermes runs on your infrastructure or a small server in your account. We configure the agent, channel, model, access boundaries, approvals, backups and cost ceiling, then test the workflow end to end.
03.
Run it beside the team
Weeks 3 and 4
The team keeps doing the task normally while the agent does the same work in parallel. We record accuracy, human edits, time saved, escalations and model cost.
04.
Keep, widen or stop
Day 30
The scorecard decides the next step. If the workflow does not justify a wider deployment, we stop. If it does, the accepted procedure becomes the baseline for the production agent.
Start with one job. Keep it only if it works.
30-day pilot
Start here
€1,900
One repeated workflow, one agent and one primary channel. Includes process mapping, installation, access rules, model setup, parallel testing, scorecard and handover.
Company deployment
For several teams
from €6,900
Multiple agents, channels and business-system connections, with private or local model options where the use case supports them, plus two weeks of close support after go-live.
Managed operations
from €690 / month
€1,490 / month for a company deployment
Tested Hermes updates, monthly cost and usage review, skill-library review, incident response and included change hours. Three-month minimum. AI usage and hosting are paid directly to the providers.
We already build the part around the agent.
MHT Experience
We built an AI proposal system for an electrical installations company that cut bidding time by 87 percent. The difficult part was not calling a model. It was document ingestion, matching, historical pricing, confidence and the operating loop around it.
DentOP
For a dental group, we built an AI surface across the clinic, including a phone receptionist, knowledge workflows and computer vision. Production meant real phone infrastructure, real CRM state and clear escalation to people.
Forward-deployed engineering
We stay inside the workflow through production instead of handing over an install script. With agent systems, permissions, procedures, model routing and edge cases become obvious only when real work starts moving through them.
Before you put an agent inside the company.
Hermes Agent is open-source software. There is no licence fee for the open-source agent itself. You still pay for the model or hardware it runs on, hosting where applicable, and the implementation and operations work around it.
Start with one repeated task.
Send us the workflow your team repeats, who does it today and which systems it touches. We will tell you whether Hermes is a good fit, whether a simpler automation would be better, and what a 30-day pilot would need.
Pilot from €1,900.