Sentry Agent Tracing Demos

Three standalone demos showing how Sentry monitors agentic behavior via AI agent tracing, one per real-world archetype: a chat bot in Slack, an assistant embedded in a product, and a headless agent in CI. Each demo is a complete, runnable app whose every model call, tool execution, and (in the storefront) database query lands as one trace in Sentry — three different frameworks, three different ways the same gen_ai.* span model gets produced.

Directory	Framework	Channel	Demonstrates
slack-agent-eve/	Eve (Vercel's agent framework)	Slack (group-order links, @mentions, DMs) + local TUI	A DoorDash group-lunch agent (via local dd-cli) traced through Eve's OTel hook + Sentry's vercelAIIntegration; a tool with its own nested LLM call; Slack threads as Sentry Conversations
storefront-commerce/	Vercel AI SDK + AI Elements on Next.js Commerce	In-app chat panel	Agent tracing next to classic app tracing: db.query spans nesting under tool spans, generative UI, Queries insights, per-session Conversations
github-harness-flue/	Flue	GitHub Action (flue run)	A multi-agent CI harness (lead + 2 parallel subagents) traced via Flue's official Sentry blueprint; logs, issues, and flue.* cross-signal tags

All LLM calls go through OpenRouter. Each demo is a self-contained npm project.

Setup (per demo)

cd <demo>
npm install
cp .env.example .env        # .env.local for storefront-commerce

Then fill in the two keys every demo needs:

• OPENROUTER_API_KEY — from openrouter.ai/keys
• SENTRY_DSN — a Sentry project's DSN (Settings → Client Keys); the storefront uses NEXT_PUBLIC_SENTRY_DSN so the browser SDK sees it too

Everything else in each .env.example is optional or channel-specific (Slack tokens, GitHub token, model overrides, source-map upload). No secrets are ever hardcoded; missing optional vars degrade gracefully.

Where to look next

• ARCHITECTURE.md — per-demo architecture diagrams, the exact call stack → Sentry span tree mapping, runbooks for generating traffic and reading the results in Sentry, and a comparison of the three instrumentation approaches.
• Each demo's own README.md — setup, run commands, and demo-specific notes.
• research/ — the four framework deep-dives (Eve, Flue, AI SDK/Commerce, Sentry agent tracing) the demos were built from: verbatim API surfaces, exact versions, gotchas. Check here before re-deriving anything.