PLG Bridge — self-serve to enterprise
An agent that watches self-serve product usage, scores accounts for enterprise readiness, and delivers an AE-ready package per account: brief, buying-committee map, draft outbound email, and a 45–60 s voice briefing.
What it does
- Ingests usage (events, seats, feature hits, API calls, plan-limit events) from streaming or batch sources, including Nexla webhooks.
- Clusters users into accounts by email domain, skipping personal-email domains, de-duping against CRM records.
- Scores every account on 5 weighted signals — seat growth, power-feature depth, admin activity, collab density, plan-limit proximity — and surfaces the top-3 contributing signals for explainability.
- Triggers autonomous research when an account crosses a configurable threshold (per-account cooldown enforced).
- Gathers public signals (funding, headcount, leadership, tech stack) via Tinyfish, structures the output with Bedrock, caches results.
- Identifies the champion from usage (admin + invites + power-feature adoption) and maps a 3–5-person buying committee.
- Drafts a ≤150-word brief + a personalized email where every sentence is grounded in a concrete usage signal.
- Produces a 45–60 s voice briefing via Vapi.
- Delivers via dashboard, Slack webhook, and a public
GET /api/briefs/:idfor CRM sync.
Every integration falls back to a deterministic mock when its credentials are missing, so the system runs end-to-end with zero external keys.
Sponsor wiring
| Sponsor | Where it's used |
|---|---|
| AWS Bedrock | src/lib/integrations/bedrock.ts — retrying client, Zod-validated JSON for committee, brief, email. |
| Nexla | src/app/api/ingest/nexla/route.ts — normalizes Nexla event payloads into our UsageEvent shape. |
| Redis | src/lib/redis.ts — accounts, users, events, scores, cooldown locks, pub/sub, activity, metrics, config. |
| Tinyfish | src/lib/integrations/tinyfish.ts — public-signal research, 24-h cache. |
| Vapi | src/lib/integrations/vapi.ts — voice brief. |
| Chainguard | Dockerfile + Dockerfile.worker — hardened base images. |
Run it locally
# 1. Install
npm install
# 2. Start Redis (or set REDIS_URL to a managed instance)
docker run -p 6379:6379 -d redis:7
# 3. Either:
# (a) click "Seed demo data" in the UI, or
# (b) load fixtures from the CLI:
npm run fixtures:generate
npm run fixtures:seed
# 4. Dashboard
npm run dev # http://localhost:3000
# 5. Async research worker (optional — the dashboard can also assemble inline)
npm run worker:research
Containers
cp .env.example .env
docker compose up --build
Smoke test
npm run smoke # seeds, simulates, forces a brief, asserts all pieces exist
Endpoints
Ingestion
POST /api/events— singleUsageEventor{ events: [...] }.POST /api/ingest/nexla— Nexla webhook format.
Accounts
GET /api/accounts— every account ranked by readiness.GET /api/accounts/:domain— full account (account, score, users, events, score history, latest brief).GET /api/accounts/:domain/activity?limit=N— agent activity log.POST /api/accounts/:domain/trigger?mode=inline|async&force=true— regenerate a brief.POST /api/accounts/:domain/simulate— demo simulator ({ intensity: "light" | "burst" | "expansion" }).
Briefs
GET /api/briefs?limit=N— recent brief packages.GET /api/briefs/:id— full brief package (for CRM sync).GET /api/briefs/:id/markdown— downloadable.mdexport.GET /api/briefs/:id/feedback— AE feedback (status + note).POST /api/briefs/:id/feedback—{ status, note }.GET /api/briefs/stream— SSE channel, pushesbrief_readyevents.
Config + admin
GET /api/config— live scoring threshold + cooldown.POST /api/config— patch ({ threshold?, cooldownDays? }).GET /api/health— Redis status + configured sponsors.GET /api/metrics— counters (briefs today, triggers today, avg time to brief).POST /api/admin/seed— one-click demo seed.POST /api/admin/trigger-all— bulk recompute + research above a min score.
Dashboard
- Home — overview stats, readiness histogram, top movers, live trigger config, filtered + searchable account table with sparkline trends.
- Account detail — score badge, plan-pressure badge, CRM badge, demo simulator, brief card (with AE status + markdown export + copy email), signals trend chart, champion, committee, research, email + voice, activity timeline, recent events, brief history, CRM record.
- Briefs — all generated brief packages, newest first.
- Activity — global agent activity feed.
- Light mode only. Live toast when any new brief lands (via SSE).
Non-functional requirements — status
- [x] Research for a single account completes in under 60 s (parallel fan-out + 24-h research cache).
- [x] Scoring updates in < 10 s from event ingest (synchronous in the ingest handler).
- [x] All LLM outputs are structured JSON validated against a Zod schema.
- [x] Containerized with Chainguard base images.
Built With
- amazon-web-services
- bedrock
- chainguard
- nexla
- redis
- tinyfish
- vapi