Concepts
An agent is a project with a personality, model choices, enabled tools, channel settings, and plan limits. It is stored in PostgreSQL and can be exported as nexus-agent.yaml.
A tool is an action the agent can call. Built-in tools cover common platform actions. Generated tools come from OpenAPI operations and run through the auth proxy. External tools use MCP servers.
Identity link
Section titled “Identity link”Identity links map a channel identity to an app user:
project_id + phone_e164 + provider_name -> provider_user_idThis lets the gateway know which app account to use when the same agent serves many users.
Auth proxy
Section titled “Auth proxy”Generated API tools call an internal proxy URL. The proxy looks up credentials, injects auth, strips untrusted auth headers, forwards the request, and filters audit data. See security.
Managed agent execution
Section titled “Managed agent execution”Every request uses the same managed General Augment agent path. Internally, General Augment keeps platform concerns such as tenant routing, credentials, RLS, billing, dashboard, and channels separate from the agent execution loop.
Usage event
Section titled “Usage event”Every message, tool call, and token event is recorded as a raw usage event and rolled into daily aggregates. Plan enforcement reads hot Redis counters first and falls back to PostgreSQL.
Continue with the architecture overview for the full request path.