Run Hermes for a team or clients
“I want to operate Hermes for a team or for paying clients, with isolation, oversight and cost attribution”
The business-ops stories converge on one pattern: isolate per client, watch everything, and be able to say what each engagement cost. Isolation comes first because it is the thing you cannot retrofit — the people running this successfully describe one profile or one instance per client from day one. Orchestration and a real dashboard follow, because selling agent work means being able to show the work. Shared memory is the subtle one: it is what lets a team of agents behave like a team, and it is also the boundary you must not accidentally cross between clients.
Real posts from the Hermes community, mirrored from Nous Research's user stories corpus.
It's right now integrated with SourceDev to index repos (self hosted), Tenderly MCP to debug onchain transactions, LLM-Wiki ingest of Litepaper of our protocol and other docs. Hopefully team will find it useful and will integrate more infra tools overtime to…
Selling AI ops to local businesses: one Hermes profile per client, fully isolated — each gets their own SOUL.md, memory, cron. Charge $497/month per client to manage their workflows. 5 clients = $2,485/month recurring. Hermes runs the work, you keep the…
If I wanted Hermes to pay for itself, I'd treat it like a junior operator, not some magic money printer. Find 25 B2B companies that could benefit from a content system, capture name/site/why-they-need-it, draft three message variants for review.
I'm on day 297 of my streak: 900,000+ seconds of compute time automated, 5,000,000,000+ tokens generated, $100,000+ in client work value automated.
My 'main agent' is my 'Chief of Staff' who has his own memory cross-project/workflow. Every 'project' (1 project = 1 Slack channel) has its own agent sub-profile with its own memory. The whole system runs on a VPS, with backup routing if the main model fails,…
- Client isolation is a hard boundary. Shared memory across agents is valuable within one engagement and a serious problem across two — scope it explicitly.
- Running someone else's business on an autonomous agent changes the risk profile entirely. Pair this with the guardrails bundle before it touches anything billable.
- Atlas has no dedicated multi-tenant billing or per-client cost-attribution project yet — today this is assembled from general telemetry plus per-client isolation.