AI agents that run real workflows.
Reduce the time your team spends on emails, documents and repeated admin. We build AI assistants and connected workflows around your tools, with human review where it matters.
Fixed scopesFrom CHF 2,500EU or on-prem
Start with a short message. We clarify the scope and cost together before any paid work.
Four practical ways to use AI.
Each engagement maps to a concrete deliverable with a fixed scope and a fixed timeline — audit, build, or retainer.
Agentic Workflows
Connect recurring tasks: route emails, prepare replies and record information in your CRM. You decide which steps run automatically and where approval is needed.
We choose the technology around your existing systems and the requirements of the workflow.
Internal AI Assistants
An internal assistant that searches your documents for answers and shows its sources. Your team can look up information directly, with unresolved questions passed on for review.
Deployed on your infra or ours. SSO, audit logs, and role-based access included.
Custom RAG & Knowledge Bases
Your documents become a queryable knowledge base. Internal wikis, support docs, contracts, research notes — ingested, embedded, and made searchable through natural language.
Stack: Postgres + pgvector or LanceDB, OpenAI / Voyage embeddings, Claude or GPT for generation.
Strategy & Build Consulting
You're not sure if AI fits your business yet? A short engagement maps your workflows, identifies the highest-ROI place to inject AI, and either we build it or hand off the plan to your team.
Outputs: written audit, prioritized roadmap, ROI estimate per use case.
What it actually looks like— annotated.
A real assistant, grounded in your data. Cited answers, a confidence threshold that escalates instead of guessing, and a per-query cost line you can defend to finance. Boring on purpose — that's what makes it ship.
What's the SLA credit if our hosted plan is down 4 hours in one billing cycle?
Per your MSA, hosted-plan downtime over 60 minutes inside a single billing cycle triggers a 10% credit[1]. A 4-hour outage falls into the major-incident tier[2], which adds 20% — total credit: 30% of that month's invoice.
Where AI actually pays off.
These tasks offer a useful starting point. Together, we assess whether the available data, effort and expected benefit make AI a suitable choice.
Customer support deflection
An assistant uses your FAQs, documents and previous tickets to prepare answers to recurring questions. Unclear requests are passed to your team.
Document & contract analysis
Upload a 200-page contract or 50 research papers; ask questions in plain language and get cited answers. Used in legal, real estate, due diligence, academic research.
Sales & lead qualification
AI reads inbound emails or form submissions, qualifies the lead, drafts a tailored reply, schedules the call. Your sales team only sees pre-qualified opportunities.
Internal knowledge search
Notion, Google Drive, Slack history, and a dozen other sources unified into one search interface. Find the answer once instead of pinging three colleagues.
Content & data extraction
Pull structured data from PDFs, scanned forms, emails, or web pages. Output clean JSON ready for your database. No more manual data entry.
Operational automations
The repetitive 30-minute tasks across your week, automated end-to-end. Email triage, report generation, status summaries, escalation routing.
The right tool, not the loudest one.
Models
Claude (Anthropic), GPT (OpenAI), open-source via vLLM or Ollama for on-premise. Embeddings: Voyage, OpenAI, BAAI/bge.
Vector storage
pgvector (Postgres-native, our default), LanceDB, Qdrant, or Pinecone — depending on scale and infra constraints.
Orchestration
Claude Agent SDK, LangGraph, custom Python or TypeScript pipelines. Tool-use, retries, cost monitoring all baked in.
Hosting
Vercel, AWS, Cloudflare Workers, or your own infrastructure. EU data residency available on request.
We run AI in production ourselves.
We ship LLM products in production
We build and operate LLM-powered products under real load and real cost constraints — our own and our clients'.
Privacy & compliance aware
Based in Zürich. We agree on hosting and access controls around your requirements. Where needed, we deploy open-source models on your own infrastructure.
Cost-conscious by default
Prompt caching, model routing (cheap for simple queries, expensive for hard ones), and observability so you know your AI bill at the end of the month.
Honest about limits
LLMs hallucinate. We design around that with citations, confidence scoring, and human-in-the-loop checkpoints where the cost of being wrong is high.
Three engagement models.
AI Strategy Audit
1-week engagement. We map your workflows, identify the top 3 AI opportunities, deliver a written roadmap with ROI estimates. Fixed price.
- Workflow audit
- Prioritized roadmap
- Build/buy recommendation
- From CHF 2,500
Fixed-Scope Build
The most common engagement. Defined deliverable, fixed timeline, fixed price. Discovery → scope → build → ship. Typical range: 3 to 8 weeks.
- Written scope & estimate
- Weekly check-ins
- Production-ready handoff
- From CHF 8,000
Ongoing Retainer
For shipped products that need ongoing iteration. Monthly block of hours, prioritized backlog, direct access. Cancel any time after the first month.
- Monthly hour block
- Prioritized backlog
- Direct Slack/Discord
- From CHF 3,500/mo
Frequently Asked Questions
Answers to the most common questions about our development services.
One call. An honest answer.
30-minute discovery call, free. We'll tell you whether AI is actually the right tool for your problem and what it would cost to build — honestly.