Skip to content

Onboarding to the Rewards Assistant Platform

Onboarding to the Rewards Assistant Platform

Section titled “Onboarding to the Rewards Assistant Platform”

The first cross-cutting onboarding guide for the Rewards Assistant — the high-level “what is this platform and how do its pieces fit together” walkthrough for a vertical team (Play, eReceipts, Restaurant Network) integrating for the first time. Repo- or spec-specific guides (“how to register a DM type via PD-3”, “how to add a new BFF enricher”) land in follow-up PRs alongside this one.

The Rewards Assistant is an agentic AI system that answers user shopping and rewards questions and proactively sends direct messages (DMs). It is not a single service — it is a platform composed of six repos that hand work to one another:

RepoRole
consumer-agentOrchestrator and user-facing surface. Hosts sub-agents, composes prompts, streams responses to mobile, persists conversation history. Owns the Chat (PC-*) and Foundation (PF-*) specs.
consumer-context-serviceDomain enrichment and BFF assembly — the context layer the agent reasons over. Owns the Services (PS-*) specs.
consumer-graph-workerDM Stack: scheduled execution, the knowledge graph, repurchase nudges. Owns the DM (PD-*) specs.
rover-agentAgent runtime.
rover-mcpMCP server exposing tools and interactive widgets to the agent.
consumer-graph-mcpMCP server over the knowledge graph.

Platform work is specified before it is built. Specs are organized into four stacks, each owned by one repo (see PF-7 §4.2):

  • Chat (PC-*) — how the agent is composed and how it executes. Start at PC-1 — Agent Composition.
  • DM (PD-*) — how proactive DMs are scheduled, delivered, and registered.
  • Services (PS-*) — the context and connector layer.
  • Foundation (PF-*) — cross-cutting platform conventions: PF-1 — Sub-Agent Lifecycle for how a sub-agent goes from dev to prod, and this guide’s parent spec PF-7 — Documentation Site for where docs live.

The Platform Index is the cross-repo table of contents; the Platform Spec Lab on Confluence is the authoritative roster (dependency graph, status, decisions).

Per PF-7, every repo publishes its docs/ directory to tech-docs.fetchrewards.com/backend/rewards-assistant/<repo>/ via the reusable docusaurus workflow. Platform specs live at docs/specs/<id>.md in their owning repo. Cross-repo links use absolute tech-docs URLs (PF-7 §4.6).

  1. Read this guide and the Platform Index.
  2. Read PC-1 to understand agent composition (skills, connectors, models) — most integrations add a skill or a connector.
  3. If you need new context data, read the Services specs and the connector-framework spec (PS-2).
  4. If you need a proactive DM, read the DM Stack specs in consumer-graph-worker.
  5. Follow PF-1 to promote your sub-agent through dev → test → prod.
  • Platform questions: the Pilot team and the Platform Spec Lab.
  • Tech-docs site issues (build failures, broken search, SSO): the Tech Docs team via #proj-tech-docs (epic DO-3576) — they own the docs platform; PF-7 is a consumer of it.