Task modelling
We define the goal, success criteria and abort behaviour — before we let the model run.
AI Agents
Autonomous agents that take repetitive work off your team — safely, inside your systems.
We build AI agents with clear tool interfaces, transparent reasoning traces and hard safety bounds — fit for regulated domains.
Reasoning loop
How an agent thinks: plan, choose tools, observe, plan again.
Every productive agent runs the same controlled loop. We make each step traceable — for engineering, for leadership, for the auditor.
Building blocks
Three building blocks no productive agent ships without.
An agent is more than a prompt. It needs a clear task, a tool inventory and hard guardrails — otherwise it becomes a liability.
Tool layer
Structured tools with validation, idempotency and rate limits — no free shell, no black-box APIs.
Guardrails
Policies, human-in-the-loop and per-run cost budgets — the agent does nothing you have not approved.
Sample trace
A real agent run — every step accountable.
This is the audit log every agent on our platform produces. Nothing happens in the dark.
support-agent · reschedule reservation
Run #2247 · 4.8 s · 6 steps- THOUGHT01
Plans approach
Guest wants to move 7:00 PM reservation to 8:30 PM. Check availability, update calendar.
- TOOL02
reservations.lookup
reservations.lookup({ guest: "M. Köhler", date: "2026-05-09" }) → reservation_id "r_8821"
- OBS03
Slot is free
Slot 20:30–22:30 is available. No conflict with other reservations.
- TOOL04
reservations.update
reservations.update({ id: "r_8821", time: "20:30" }) → ok
- TOOL05
mail.send
mail.send({ to: "guest", template: "reschedule_confirmation" }) → queued
- FINAL06
Reply to guest
"Your reservation has been moved to 8:30 PM. A confirmation is on its way."
Agent library
Six battle-tested agent patterns we have shipped to clients.
Each pattern ships with tool definitions, evaluation sets and monitoring — ready to be tuned to your domain.
Live
Support agent
Tier-1 service
Resolves recurring tickets, escalates with context, and logs every action cleanly into your CRM.
crm.lookupkb.searchmail.send
Live
Research agent
Knowledge work
Searches your knowledge base, summarises findings, and cites sources with every sentence.
kb.searchweb.searchdoc.summarize
Beta
Inbox agent
Mail triage
Sorts incoming mail, drafts replies and creates tasks in your tracker — humans approve.
mail.readtasks.createmail.draft
Live
Ops agent
Operations
Watches dashboards, alerts on thresholds and triggers standard runbooks without touching production flows.
metrics.readpager.notifyrunbook.run
Live
Task agent
Project management
Keeps backlogs clean, closes tickets when done and surfaces open items before every stand-up.
tasks.listtasks.updatestandup.brief
Beta
Analytics agent
BI
Answers business questions against your BI layer in natural language — with the underlying SQL exposed.
db.querychart.rendermetric.define
By the numbers
This is hafencity.dev
15+ Experts
Engineers, designers, and strategists working as one practice from our Hamburg HQ.
50+ Clients
Companies across consumer, healthcare, and B2B trust us with their digital products. Long-term partnerships are the default.
97% Recommendation rate
Repeat engagements and references our buyers actually call. Trust compounds when delivery does.
50+ Launches
Custom mobile and web products shipped from concept to maintenance — owned end-to-end by our team.
100% In-house
Strategy, design, and engineering all live at our Hamburg HQ. One team, one project lead, accountable to you from kickoff to launch.
Since 2023
Helping companies ship digital products — and growing alongside the teams we work with.
Next steps
Let's talk about your project
Book a 30-minute discovery call. We'll review your goals, surface unknowns, and outline how we would run the engagement.