Learning Guide
A guided path through the entire system. Each stage pairs doing (a hands-on tutorial or exercise) with understanding (the concept pages that explain what you just saw). Finish all five stages and you can maintain and extend every part of this platform.
How the system works — the 60-second mental model
- An agent is configuration — prompt + language + tone + tool allow-list, stored as a row (Agents).
- Endpoints animate the agent — chat wraps its prompt around your message; email endpoints run extraction pipelines; document endpoints search its knowledge base (API Reference).
- Services own all intelligence — every OpenAI call, prompt, chunk, and vector lives in five service modules (Backend).
- Everything is recorded — runs, tool calls, workflow steps, with latency and cost estimates (Observability).
- The frontend renders the records — server components fetch fresh data; one form writes through a proxy (Frontend).
Everything else in these docs is detail on one of those five sentences.
Stage 0 — Run it
Do: Installation → Quick Start. Understand: Project Structure · Configuration. Checkpoint: both servers up; you created an agent and got a chat reply.
Stage 1 — The request path
Do: Your First Agent — build an agent deliberately and trace every side effect.
Understand: System Overview · Request Lifecycle · Database Layer · ORM Models.
Checkpoint: you can narrate a chat request from curl to committed rows without looking.
Stage 2 — Multi-agent workflows
Do: Email Workflow Tutorial — run the pipeline, then read its trace like a flight recorder. Understand: Workflows · Email Workflow · Service: Multi-Agent · Prompt Catalog · Evaluation. Checkpoint: you can explain why the reviewer runs after task commit, and what happens on stage-3 failure.
Stage 3 — RAG
Do: RAG Tutorial — ingest, inspect chunks, watch the graph answer and refuse. Understand: RAG Overview · Ingestion · Retrieval · Corrective RAG · Service: RAG Graph. Checkpoint: given a wrong RAG answer, you know whether to blame chunking, retrieval, the evaluator, or the prompt — and how to check each.
Stage 4 — The frontend
Do: change the dashboard to show a metric it currently hides (e.g., failed workflow runs as a StatCard) — tokens only.
Understand: Frontend Overview · Pages · Components · API Proxy · Data Fetching · Design System.
Checkpoint: you know when a fetch belongs in a server component vs. behind a proxy route, and why.
Stage 5 — Extend it
Do: all three Guides — endpoint, tool, migration — ideally as one feature (e.g., the sentiment tool end to end). Understand: Design Decisions · Coding Standards · Known Issues (fixing one is the graduation exercise). Checkpoint: you shipped a change that respects all five contracts.