Ty
48566993f6
Product layout for Lumina / Salon_Assistant at 0.1.0-design. No product implementation until explicit build.
27 lines
1.7 KiB
Markdown
27 lines
1.7 KiB
Markdown
# Platform research notes (NemoClaw / Hermes)
|
|
|
|
Normative links for implementers. Prefer platform-native operations over bespoke control planes.
|
|
|
|
## NemoClaw
|
|
|
|
- [Hermes architecture](https://docs.nvidia.com/nemoclaw/latest/user-guide/hermes/reference/architecture.md) — CLI, blueprint, OpenShell gateway, sandbox, L7 credential injection
|
|
- [Understand Runtime Changes](https://docs.nvidia.com/nemoclaw/latest/user-guide/hermes/manage-sandboxes/configure-sandboxes/understand-runtime-changes.md) — what hot-reloads vs rebuild
|
|
- [Manage Messaging Channels](https://docs.nvidia.com/nemoclaw/latest/user-guide/hermes/manage-sandboxes/messaging-channels/manage-messaging-channels)
|
|
|
|
Key host commands: `nemohermes onboard`, `inference set`, `config set`, `policy-add`/`policy-remove`, `channels add`, `snapshot create`, `rebuild`, `shields up/down`, `gateway restart`.
|
|
|
|
In-sandbox source of truth: `/sandbox/.hermes/config.yaml` + `.env` (trust-anchored; do not hand-edit expecting trust).
|
|
|
|
## Hermes
|
|
|
|
- [Security](https://hermes-agent.nousresearch.com/docs/user-guide/security) — write safety, approvals, gateway allowlists, MCP env filtering
|
|
- [Configuration](https://hermes-agent.nousresearch.com/docs/user-guide/configuration)
|
|
- [MCP config](https://hermes-agent.nousresearch.com/docs/reference/mcp-config-reference) — `tools.include` / `exclude`
|
|
- [Configuring Models](https://hermes-agent.nousresearch.com/docs/user-guide/configuring-models) — main + auxiliary (vision)
|
|
|
|
File write safety blocks credential/`config.yaml` self-edits via agent tools. Mutations belong on the **host** via NemoClaw sealed commands.
|
|
|
|
## Decision
|
|
|
|
No parallel control API. Product scripts wrap `nemohermes` / `openshell` only.
|