# HermesCo

> A one-agent company: Hermes decides, NVIDIA Nemotron screens, Stripe settles, with every dollar gated by a human-in-the-loop Treasury.

## Docs

- [Run an agent turn](https://docs.hermesco.ai/api-reference/agent.md): POST /api/hermesco/agent
- [One machine: status, actions, destroy](https://docs.hermesco.ai/api-reference/agent-machine.md): GET, POST, and DELETE /api/hermesco/agents/[id]
- [Fleet: list and provision](https://docs.hermesco.ai/api-reference/agents.md): GET and POST /api/hermesco/agents
- [API overview](https://docs.hermesco.ai/api-reference/introduction.md): Every HermesCo endpoint under /api/hermesco, with conventions for workspaces, errors, and runtime.
- [Read the Treasury](https://docs.hermesco.ai/api-reference/treasury.md): GET /api/hermesco/treasury
- [Approve or deny a proposal](https://docs.hermesco.ai/api-reference/treasury-decide.md): POST /api/hermesco/treasury/decide
- [Deposit capital](https://docs.hermesco.ai/api-reference/treasury-deposit.md): POST and GET /api/hermesco/treasury/deposit
- [Reset a workspace](https://docs.hermesco.ai/api-reference/treasury-reset.md): POST /api/hermesco/treasury/reset
- [Architecture](https://docs.hermesco.ai/architecture.md): How the brain, safety screen, money rail, agent body, and persistence fit together.
- [Identity and access](https://docs.hermesco.ai/concepts/auth.md): Guest-first by default, with an optional Google sign-in for a named operator on the approval ledger.
- [Per-agent Fly machines](https://docs.hermesco.ai/concepts/fly-machines.md): Each agent gets its own dedicated, multi-core Fly Machine as its body. Real work, metered, suspends to $0 when idle.
- [The Job Desk](https://docs.hermesco.ai/concepts/jobs.md): HermesCo sells real units of compute work. A customer brings a brief, pays a Stripe link, and the agent runs the job on its own machine and returns the deliverable.
- [Persistence](https://docs.hermesco.ai/concepts/persistence.md): Convex when configured, an in-memory store otherwise. The Treasury survives across requests either way.
- [The pipeline](https://docs.hermesco.ai/concepts/pipeline.md): Hermes decides, NVIDIA Nemotron screens, Stripe settles. One agent, one loop, three systems.
- [Safety screen](https://docs.hermesco.ai/concepts/safety.md): NemoClaw: a deterministic policy gate layered with NVIDIA Nemotron classification, so the safety outcome never depends on a model being available.
- [Treasury](https://docs.hermesco.ai/concepts/treasury.md): The human-in-the-loop money control. Real capital, hard caps enforced in code, a signed ledger, and zero fabricated revenue.
- [Configuration](https://docs.hermesco.ai/configuration.md): Every environment variable HermesCo reads, what it controls, and its default.
- [Credits](https://docs.hermesco.ai/credits.md): Every service HermesCo actually runs on, and exactly what each one does.
- [Deployment](https://docs.hermesco.ai/deployment.md): How HermesCo runs in production on Fly, and the secrets that turn on each capability.
- [HermesCo](https://docs.hermesco.ai/introduction.md): A one-agent company that earns, spends, and runs real operations, with every dollar gated by a human-in-the-loop Treasury and hard caps it cannot breach.
- [Quickstart](https://docs.hermesco.ai/quickstart.md): Run HermesCo locally, drive the agent, and watch the Treasury enforce its caps.

## OpenAPI Specs

- [openapi](https://docs.hermesco.ai/api-reference/openapi.json)