embr-pulse

📍 Team-shared reference copy. Live deployment, end-to-end agent receipts (the Copilot-authored PRs, issue #8 → PR #9, incident #139), and active development happen at seligj95/embr-pulse. This repo mirrors the source + docs so anyone in embr-devs can read, fork, or adapt without dependency on a personal account. Workflows here are inert (renamed to .yml.example) — there's no live cron, no live monitor.

---

Team feedback aggregator hosted on Embr — every part of its lifecycle (development, deployment, monitoring, fixes) is driven by agents.

Reference architecture for coreai-microsoft/embr#374.

What this is

A real internal app the Embr team uses to give feedback. Each submission becomes a GitHub issue. GitHub Copilot's coding agent picks up issues, opens PRs, and Embr auto-deploys them. A Foundry monitor agent watches production telemetry and opens incident issues when SLOs trip.

The app's purpose is the agent loop. Building it surfaces what Embr needs to make this experience first-class for customers.

Status: All three loops are operational. Receipts:

• Loop 1 (code): every PR in this repo merged via Copilot is a Loop 1 receipt.
• Loop 2 (feedback → ship): issue #8 → PR #9 in ~25 minutes.
• Loop 3 (self-heal): incident #139 — Foundry monitor agent detected a synthetic triage-failure spike, filed a real incident issue with hypothesis + suggested action, and deduplicated subsequent triggers onto the same issue.

What we learned

Building this surfaced 15 platform-friction points and one architectural opinion worth surfacing up front:

Don't make Embr "an agent platform." Make Embr the platform that agent-driven apps run on.

Triage prompts, monitor classifiers, and Copilot routing are per-customer code — they belong in the app. Workload identity, scheduled jobs, deployment-lifecycle webhooks, and observability auto-wiring are per-platform — they belong in Embr.

Full reasoning: docs/embed-in-embr.md. Upstream impact: 15 gaps lodged against coreai-microsoft/embr (11 new issues + 4 comments on existing tracking issues). Catalog: docs/platform-gaps.md.

Docs

Start here (15-min read for anyone landing cold):

• docs/customer-story.md — narrative pitch with real receipts (issue #8 → PR #9 in ~25 minutes)
• docs/embed-in-embr.md — standalone agent vs ADC-embedded: our recommendation and why

Customer-facing:

• docs/reference-architecture.md — what we built, distilled for customers who want to copy it
• docs/demo-script.md — canonical happy-path demo we engineer toward

Engineering:

• docs/design.md — full architecture and rationale
• docs/platform-gaps.md — running log of Embr platform improvements (with filed-as links)

Local development

# 1. Install
npm install

# 2. Spin up Postgres (any way you like — Docker, Postgres.app, etc.)
#    Then export DATABASE_URL or copy .env.example to .env.local and edit.
cp .env.example .env.local

# 3. Apply schema
npm run db:migrate

# 4. Run dev server
npm run dev
# → http://localhost:3000

Deploy to Embr

# One-time setup (push to seligj95/embr-pulse, link via Embr GitHub App):
embr quickstart deploy seligj95/embr-pulse -i <installation_id>

After that, every push to main auto-deploys.

Stack

Layer	Tech
App	Next.js 15 (App Router) + TypeScript
DB	Postgres (Embr-managed)
Auth	NextAuth + Microsoft Entra ID
Coding agent	GitHub Copilot
Triage / Monitor	Microsoft Foundry agents
Telemetry	App Insights + existing Embr Kusto
Hosting	Embr