Inbound qualification → CRM
Stripe-style intake form posts to n8n. Claude classifies, scores, and drafts a reply. Hubspot deal created with full context.
AI workflows your team will actually ship.
n8n + Claude Code, EU-hosted infrastructure, 30-day fit-check exit on every build.
{{ TODO: copy }} - client logos OR one-sentence testimonial
01 / workflows
Workflows, shipped end-to-end
Support triage with private context
Inbox webhook → embeddings over your docs + past tickets → Claude writes a draft reply, agent approves in one click.
Lead enrichment from a single domain
Domain in, structured firmographics out: ICP fit, tech stack, hiring signals, latest news, contact suggestions.
try the demo ›
Weekly revenue digest
Pulls Stripe, Hubspot, Postgres. Claude writes a 6-line summary with anomalies flagged. Posted to Slack Monday 09:00.
Churn-risk alerts
Usage + support sentiment + invoice lag → risk score. Anything above threshold lands in CS lead's queue with reasoning.
RFP / security questionnaire autofill
Upload PDF/DOCX. Workflow extracts questions, retrieves your prior answers, drafts new ones, exports back to original format.
02 / case study
From zero posts to daily drafts on a Shopify store
In-house pilot, Lithuanian e-commerce brand
Posts/week ~0 → 2-3Time per post: hours of blank-page → up to 15 min of proof and polish.
Problem
One-person store. Social was supposed to drive traffic but wasn't happening. The slow part wasn't typing, it was deciding what to post and which product to tie it to before writing started.
Solution
Daily n8n run pulls Lithuanian calendar context (date, name day, holiday, season, weather), fetches in-stock Shopify products, asks Claude for 2-3 drafts per day. Some tied to the calendar, some to a specific product. Tracking links built at draft time.
- Build time
- ~25 engineer-hours
- Stack
- n8n · Claude · Shopify
03 / about
Who you're hiring
Erikas. Vilnius. AI Enablement Lead at a Lithuanian data infra company by day. Joined writing Python, ended up running a team of engineers. Got a US patent along the way. The thing that keeps me in this work: most office tasks are wasteful and nobody questions it because that's how it's always been done. Status reports rewritten when last week's would do. Same data copy-pasted across three tools. Automate the boring parts and people get their afternoons back for work that actually needs thinking.
LinkedIn04 / how it works
Three ways to start
- 01
Audit
Map your highest-leverage automations. Walk away with a prioritized backlog.
- 02
Build
Pick a workflow from the audit. I ship it to production in 2–4 weeks.
- 03
Retainer
Ongoing maintenance, tuning, and new builds. Month-to-month after a 90-day initial term.
05 / pricing
Engagement tiers
Audit
€3,500/ 10 working days
Map your highest-leverage automations. Walk away with a prioritized backlog.
- ›Stack + data audit
- ›Top 5 workflows scored by ROI
- ›Build estimates per workflow
- ›60-min walkthrough call
Build
from €11,000fixed scope
Pick a workflow from the audit. I ship it to production in 2–4 weeks.
- ›Fixed scope, fixed price
- ›Production deploy on your infra
- ›Monitoring + rollback included
- ›Loom handover + README
- ›30-day fit-check exit on the first month
most common
Operations Tier
from €2,200/ month
Ongoing maintenance, tuning, and new builds. Month-to-month after a 90-day initial term.
- ›2 workflows under maintenance
- ›≤30h of new build per quarter
- ›Biweekly office hour
- ›Next-business-day SLA, 4h on critical incidents
Operations Tier - Plus
€3,500/ month
Deeper engagement when 2 workflows aren't enough.
- ›4 workflows under maintenance
- ›≤50h of new build per quarter
- ›Weekly office hour
- ›4h response SLA, business hours
06 / stack
Boring, EU-hosted, all yours
- Workflow engine
- n8n, self-hosted on a VM you own. Pinned versions, no per-seat tax. Workflow definitions live in your git org.
- LLM providers
- Anthropic, OpenAI, Mistral, OpenRouter as fallback. EU endpoints where offered. Model choice is per-workflow - swap in minutes.
- Glue code
- Python by default. TypeScript when the integration is JS-native.
- Storage & state
- Postgres 16 alongside n8n. State, logs, pgvector embeddings. Supabase EU if you'd rather not self-host.
- Observability
- Every LLM call and workflow run traced through Langfuse - inputs, outputs, tokens, latency, failures. Source for the monthly report.
- Infra
- Hetzner (Falkenstein or Helsinki), Caddy + auto Let's Encrypt, UFW + Cloud Firewall, nightly encrypted Postgres backups to Object Storage. nginx on request.
Hetzner and LLM providers bill you directly - VM €10–20/mo typical, model spend on your own API keys. Both separate from the retainer. Tracing runs on our side: monthly Langfuse report on call volume, tokens, latency, failures and fixes. Not another login.
07 / faq
Questions worth asking up front
- How is pricing structured?
- Four tiers. Audit Sprint: €3,500, 10 working days, fixed scope. Build: from €11,000, fixed scope, fixed deadline; multi-workflow or LLM-heavy builds quoted up to €16,000. Operations Tier: from €2,200/month, 90-day initial term then month-to-month. Operations Tier - Plus: €3,500/month for teams that need 4 workflows under maintenance and weekly cadence.
- Who owns the IP?
- You do. All workflows, code, prompts, and infrastructure are yours from day one. Repo lives in your GitHub org. No vendor lock-in clauses, ever.
- Where does the data live?
- EU-resident by default. Hetzner Nuremberg or Helsinki, Supabase EU regions, Anthropic EU endpoint where available. DPA on request. No data leaves the EU without your explicit approval.
- What does the refund policy look like?
- Build: 30-day fit-check exit. If, during the first 30 days of the build, either side determines it's not a fit (scope drift, surprise integration, cadence mismatch), you exit. Build fee refunded minus hours already worked at a published time-and-materials rate. Audit: refundable for the first 3 working days, no questions. Retainer: month-to-month after a 90-day initial term - no penalty to stop after that.
- What do I own after the engagement?
- Source for every workflow, deployment scripts, n8n instance, database, secrets. A README explaining how each piece works. A Loom walkthrough for non-engineers on your team.
- What happens if a workflow breaks at 3am?
- Operations Tier: next-business-day SLA on non-critical issues, 4h on critical incidents - all during EU business hours. Operations Tier - Plus: 4h SLA on everything during EU business hours. Outside hours on both tiers, alerts route to your team. Every workflow ships with health checks and a documented rollback procedure.
- What does onboarding look like?
- Audit: read-only access on day 1, scored backlog and Notion doc shipped on day 10. Build: kickoff within 7 working days of signature, working prototype by day 5, production deploy in 2–4 weeks. You'll have artifacts in your hands fast either way.
- What's in scope for the retainer?
- Operations Tier: 2 workflows under maintenance, ≤30h of new build per quarter, monitoring, prompt tuning, model upgrades, biweekly office hour. Operations Tier - Plus adds 2 more workflows under maintenance, ≤50h quarterly build, and a weekly office hour. Out of scope on both: full-team training, building a separate product, anything that would need a second engineer.
- How do you keep an eye on workflows after they ship?
- Every LLM call is traced on amberstack's observability stack. You get a monthly report - call volume, token spend, latency, failures and fixes. No dashboard for you to babysit.
08 / contact
Book a 30-minute audit call
Pick a slot below or email hello@amberstack.eu. Calls are 30 minutes, no slide deck, screen-share recommended so I can poke at your stack live.