Services
AI services, from first pilot to production
Service 01
AI customer assistants
Website chat, e-mail and phone assistants that answer from your real data — orders, bookings, policies — and bring in a person when they should.
What we deliver
- An assistant for your website, help centre, inbox, messaging channels or phone line, in your tone and languages.
- Answers grounded in your knowledge base and live records, with each customer seeing only their own data.
- A clear handoff to your team, with the conversation and context attached.
- Leads and tickets created in the tools you already use.
Typical examples
- “Where is my order?” answered from shipment data
- First replies to support e-mails — or silence when your documents don’t hold the answer
- An after-hours phone line that answers, transfers or books a call-back
Already running in our products
- Website assistant grounded in live dataPexpy · Claude, GPT, Own model
- Emailbot that is grounded or silentPexpy · OpenAI, Anthropic, Embeddings
- AI phone agentPexpy · Telephony, Speech-to-text, LLM
Service 02
AI workflow automation
Hand AI the routine reading, sorting and drafting — with an approval step wherever a mistake would matter.
What we deliver
- Extraction from documents, receipts and invoices into fields your systems accept.
- Classification and routing of e-mails, tickets and leads.
- Drafts of replies, quotes and summaries for a person to approve.
- Connections to your CRM, ERP or store, to Zapier, Make or n8n, or to your own APIs.
Typical examples
- Expense receipts read into vendor, tax number, VAT and totals
- Sales calls turned into summaries, action items and follow-up tasks
- Inbound requests sorted and routed to the right person
Already running in our products
- Call transcription and analysisPexpy · OpenAI Whisper, Deepgram, LLM
- Receipt and invoice readingPexpy · GPT vision, Claude vision
Service 03
Knowledge assistants & search
Search and answer across your documents, help centre and internal knowledge — with sources shown and access rules respected.
What we deliver
- Retrieval over your documents, help articles and records (RAG), kept in sync as they change.
- Answers that cite their sources, and that say so when the sources don’t cover a question.
- Access rules applied before retrieval, so people only get answers from what they may see.
- Quality checks on real questions before and after launch.
Typical examples
- A public help centre with semantic search and AI answers
- An internal assistant for policies, price lists and product specs
- Answers in the reader’s language from documents written in another
Already running in our products
- Emailbot that is grounded or silentPexpy · OpenAI, Anthropic, Embeddings
- Help-centre search with cited answersPexpy · Embeddings, GPT, Claude
Service 04
AI enablement for teams
Roll out ChatGPT Business or Enterprise, or Claude Team or Enterprise, so your people use AI well — and safely.
What we deliver
- Use-case discovery with each team, and a short list of where to start.
- Workspace setup: access, data controls, connectors and shared projects or custom GPTs.
- Usage guidelines your people can actually follow, and hands-on training.
- Connections to your own systems through MCP servers and APIs.
Typical examples
- A sales team working from shared prompts and account research projects
- An MCP connector that lets the assistant read your CRM within each person’s permissions
- A one-page AI usage policy with clear do’s and don’ts
Already running in our products
- MCP server for AI clientsPexpy · Model Context Protocol
Service 05
Custom AI development
AI features inside your own product: copilots, agents that use tools behind approval steps, MCP servers and multi-model engines.
What we deliver
- Copilots and assistants built into your web app, admin panel or customer portal.
- Agents that call your APIs through scoped tools, with approval gates and an audit trail.
- A model layer with provider choice, bring-your-own-key, fallbacks and usage metering.
- Evaluation sets, structured and validated output, and clean handover to your team.
Typical examples
- A business copilot that reads dashboards and suggests next steps
- Per-customer AI usage limits and billing in a SaaS product
- An MCP server that exposes your product to Claude, ChatGPT and other clients
Already running in our products
- Website assistant grounded in live dataPexpy · Claude, GPT, Own model
- MCP server for AI clientsPexpy · Model Context Protocol
- Incident explanationsLaunchOps · Claude
Service 06
AI security & operations
Keep AI in production safe, affordable and reliable: reviews, guardrails, monitoring and model upgrades.
What we deliver
- A review of your AI features: data flows, key handling, prompt-injection and tool risks.
- Guardrails: rate limits, quotas and spend caps, input and output validation, safe outbound requests.
- Monitoring of quality, cost and errors, with logs that leave secrets out.
- Planned model upgrades and provider fallbacks, so a change upstream doesn’t surprise you.
Typical examples
- Switching to the cheapest adequate model as a monthly budget runs down
- Blocking AI-triggered requests to private networks (SSRF)
- An audit trail of which tools an agent called — names, not values
Already running in our products
- MCP server for AI clientsPexpy · Model Context Protocol
- Incident explanationsLaunchOps · Claude
How we work
Start small, prove it, then scale what works
Every engagement starts with a clear scope and agreed success measures, and you decide after each stage whether to go further.
- 1
Discovery workshop
A focused look at your processes, data and risks. You leave with a short list of use cases, ranked by value and effort, and a pilot plan.
Prioritised roadmap and pilot scope
- 2
Pilot
One use case on your real data, with success measures agreed up front. You see the result working before you commit to more.
A working pilot and a measured result
- 3
Build & integrate
We take what worked into production: integrations, guardrails, approval steps, monitoring, documentation and handover.
AI in production, documented
- 4
Run & improve
Ongoing care: quality and cost reports, model updates, new use cases and support when something needs attention.
A partner after launch
The same four steps — discover, pilot, build & integrate, run & improve — apply whether we build for you, with your team, or for your clients as a partner.
Who we work with
Built for teams that want AI working, not just discussed
Growing businesses
Teams that want AI to answer customers, cut routine work and surface what matters — without hiring an AI team first.
Agencies
Agencies that want to offer AI assistants and automation to their clients, under their own brand or alongside us.
Software companies
Product teams adding copilots, agents or MCP connectors to their software, and running them safely at scale.
Not sure where AI fits yet?
Tell us how your team works today. We will suggest one or two places where AI would pay off first — and what a pilot would look like.