Module execution 

Agent execution service — single place handling conversation history loading, memory injection, tool coordination, observability, usage/cost, token budget, and loop detection.

Structs

AgentRequest

Request for unified agent execution.

Execute

Unified agent execution service — the single place handling:

ExecutionResult

ModelOverride

Canonical per-request model override used by the LLM interceptor.

Tools

Tenant-aware tool capability.

Enums

AgentSource

Result of Execute::run including resolution metadata.

Traits

RunTracker

Trait for tracking active agent runs. Implemented by the root crate’s ActiveRuns and injected into Execute via the Context.

Functions

request_tenant_ctx

Request-path tenant: open the realm (or isolate) then intercept TenantContext. Background jobs keep using tenant_scope (isolate only, no intercept).

request_user_scope

JWT user: isolate when no tenant is present. Does not invent TenantContext.

tenant_scope

Derive tenant for execute without requiring the postgres-only resolver module. Scope tools and execution to one tenant. Isolate wins over intercept.

user_id_from_ctx

Derive user/tenant scope: Execute isolate label (strip tenant:/user:), then TenantContext intercept, then fallback.