Talk to your business. Hermes hears intent, not commands. It schedules. It summarizes. It drafts. It acts — across every module, with the same permissions and audit trail you have.
Hermes hears intent in plain language and turns it into the right action — with full context from your CRM, calendar, deadlines, and accounting.
Talk to it from your phone or browser. It transcribes, understands intent, and confirms before acting.
Reads contacts, projects, invoices, calendar. Creates records, sends messages, schedules events.
Each org's Hermes is grounded only in that tenant's data, with full RLS isolation. Never cross-org.
Anthropic, OpenAI, or your own local model. EU-hosted inference for sensitive deployments.
Hermes is engineered for ERP semantics — not for prompt-engineered improvisation.
Every Hermes action is logged with the prompt, the model used, the tools invoked, and the resulting database writes. Investigate any decision after the fact.
Hermes runs under your role. It can't do anything you can't do. Field-level ACLs, record-level rules, organization scoping — all enforced.
Modules expose typed tools the agent calls — not free-text prompts. Type-safe, predictable, idempotent.
pgvector embeddings of your records, indexed per tenant. The agent retrieves before it answers — no hallucinations from training.
Run on Anthropic's EU endpoints, OpenAI's EU region, or your own local llama.cpp / Ollama / vLLM. Sensitive data never leaves the perimeter.
Swap providers without code changes via the LLM adapter. Add custom tools to expose plugin-specific actions to the agent.
Apply for partner access and we'll spin up a sandbox tenant with Hermes pre-wired. Five minutes to your first voice command.