Deep dives

Start with the system overview to understand who decides, who executes and who records history. Then choose the owner or failure path you need. These pages describe source; they are not a live deployment inventory or evidence that every planned feature is accepted.

The architecture pages explain component boundaries. The reference records contracts and endpoints. These deep dives connect those boundaries to request flow, state, and failure handling.

How the views fit together

The system overview uses arc42's twelve complementary views: goals, constraints, context, strategy, building blocks, runtime, deployment, crosscutting concepts, decisions, quality requirements, risks and glossary. Subsystem pages combine the opening context views and delegate shared deployment/decisions to canonical pages rather than repeat a framework. Each expands its important runtime, failure and recovery paths.

Static dependencies do not establish message order. Runtime sequences do not establish a deployed topology. Quality scenarios name an observable response and verification boundary, not an invented production SLO. ADRs retain their actual status in the system decision view.

System and orchestration

• System overview — Trace requests, identify state owners, and distinguish history from execution state.
• Gateway — Follow authentication, allowlisted routing, browser access, and journal-backed event streams.
• Orchestration — Understand decision gates, validated WorkPlans, explicit dispatch, and recovery.
• Events and sessions — Learn how the journal, addressed messages, session material, and usage records work.
• Agent runtime — Follow runtime bootstrap, SDK turns, tool authorization, and recovery.
• Retained MCP compatibility adapter — Understand source OpenAPI-to-tool translation, not a supported product entrypoint.

Identity, configuration, and execution resources

• Identity and secrets — Understand broker identity, purpose-bound grants, secret references, and credential redemption.
• Projects and configuration — Learn how project authority, revisioned settings, skills, and accepted run selections work.
• Environment and sandbox — Follow environment lifecycle, workspace volumes, network egress, and runtime placement.
• Providers and models — Understand provider selection, capability negotiation, resource pins, and model bindings.
• Source Control — Follow repository pins, isolated workspaces, output capture, pull requests, and approval-bound merges.

Data and operations

• Knowledge and memory — Learn how records, proposals, accepted decisions, context compilation, and Memory providers work.
• Persistence — Understand service-owned PostgreSQL schemas, object bytes, outboxes, inboxes, and retry boundaries.
• Telemetry — Learn how OpenTelemetry signals and exporters relate to durable product evidence.
• Infrastructure and deployment — Review Azure and Kubernetes source, workload configuration, images, and release boundaries.
• Testing strategy — Choose checks for service logic, PostgreSQL behavior, harnesses, and acceptance evidence.