AI agent development company UK:
agents that do the actual work.
Magora is a London AI agent development company for UK businesses. We build custom AI agents that read documents, work across your CRM and ERP, and hand decisions to people when they should. Free working prototype in 5–10 days, AI agents from £20,000.
AI agents at Magora,
at a glance.
The short answer on who builds our agents, what they cover, what they cost and how long they take. For wider AI work, see our AI development company UK page.
- Who
- Magora is a London AI agent development company, building software since 2010.
- What we build
- Document agents, workflow and process agents, customer service and chat agents, internal knowledge copilots (RAG), real-time and voice assistants, and AI inside your existing product.
- From what price
- 2026 prices in GBP: working prototype free; Discovery from £3,000; AI proof of concept from £15,000; an agent automating one internal workflow from £20,000; knowledge agent (RAG) from £30,000; production LLM application from £40,000. Adding AI to an existing application typically costs £20,000–£80,000; support is from £1,000 a month. See UK software development pricing, fixed-price software development and the cost calculator.
- How long
- Prototype in 5–10 working days and Discovery in two weeks. A production agent for a single workflow typically ships in 6–8 weeks after Discovery; multi-agent systems run 3–6 months.
- Safety and data
- Narrowest permissions that do the job, an audit log of every action, human approval where a mistake would be costly, and evaluations on every change. Data stays in UK or EU regions; source code, prompts and evaluation sets transfer to you on payment.
- Updated
- Updated: 13 October 2026
AI agent development
services.
An AI agent is software that uses a large language model to decide what to do next, calls your tools and systems to do it, and checks the result before moving on. A chatbot answers; an agent acts. These are the six kinds of agent UK businesses ask us for most.
Document agents
Contracts, invoices, claims, reports and application forms read, checked against your rules, summarised and filed — with the source line quoted beside every extracted value.
Workflow & process agents
Multi-step work that crosses systems: reconcile the CRM with the ERP, raise the ticket, update the record, chase the missing field. Built on the same integrations your team already trusts.
Customer service & chat agents
AI chatbot development that goes past FAQs: agents that look up the order, change the booking and escalate to a person with the whole conversation attached.
Internal knowledge copilots
Retrieval-augmented (RAG) assistants over your policies, tickets and manuals, answering with citations — so staff can check the answer instead of trusting it.
Real-time & voice assistants
Agents that listen alongside your people and prompt them live — as in Empath AI, which guides call-centre agents inside Microsoft Teams.
AI inside your product
Generative AI features added to the app or platform you already run — search, summaries, drafting, triage — without rebuilding what works. Broader ML and model work lives on our AI & ML development page.
AI agent, chatbot, RAG
or workflow automation?
The right answer is often the simplest tool that does the job. This is how the four compare; the free prototype and Discovery show which one your workflow needs.
| Chatbot | RAG assistant | AI agent | Rule-based workflow automation | |
|---|---|---|---|---|
| What it does | Answers questions from a script or the model's general knowledge | Answers from your own documents and cites the source | Plans the steps, calls your systems, checks the result and hands over to a person when it should | Follows fixed if-then rules between systems |
| Best for | Simple FAQs and first-line questions | Accurate answers from private or changing content | Multi-step work that varies from case to case | Stable, predictable steps that rarely change |
| Acts in your systems? | No | No: it reads and answers | Yes, within the permissions you grant | Yes, only the steps written in advance |
| Magora starting point | Scoped after the free prototype | Knowledge agent (RAG) from £30,000 | From £20,000 for one internal workflow | If rules are all you need, we say so |
Use cases by
business function.
High-volume work that people do by reading, re-typing and chasing is where agents earn their cost first.
| Function | Typical tasks | Agent type |
|---|---|---|
| Finance and back office | Invoices, claims, contracts and forms read, checked against your rules, summarised and filed, with the source line quoted beside each value | Document agents |
| Customer service | Look up the order, change the booking and escalate to a person with the whole conversation attached | Customer service and chat agents |
| Sales and operations | Reconcile the CRM with the ERP, raise the ticket, update the record, chase the missing field | Workflow and process agents |
| People, IT and internal queries | Staff questions answered from policies, tickets and manuals, with citations so the answer can be checked | Knowledge copilots (RAG) |
| Contact centres | Live prompts for people during a call, as in Empath AI, which guides call-centre agents inside Microsoft Teams | Real-time and voice assistants |
Small business? You do not need a large budget to find out whether an agent is worth building: the working prototype is free and a Discovery sprint starts from £3,000. Read the practical guide what AI agents work best for small business automation and our checklist is your business ready for an internal AI agent?.
How we build agents
you can trust.
Same senior engineers from the first workshop to handover. Weekly demos. No offshore rotation.
One workflow, measured
We start with a single workflow and time it as it runs today. The agent is judged against that baseline — hours saved, error rate, turnaround — not against a demo.
Guardrails and a human in the loop
Agents get the narrowest permissions that do the job, an audit log of every action, and a person's sign-off wherever a mistake would cost money, a customer or a regulator's attention.
Model-agnostic, UK data
OpenAI, Anthropic Claude, Google Gemini or open-source models, chosen on accuracy, cost and privacy. Data stays in UK or EU regions, and swapping the model later does not mean rebuilding the agent.
How much an agent
is allowed to do.
Autonomy is a design decision, agreed and written down before the build and widened only as the evaluation results justify.
Suggests, a person acts
The agent drafts the answer, the record update or the email, and a person approves and sends it. A safe starting point for any new workflow.
Acts on low-risk steps
Routine, reversible steps run on their own and every action is written to the audit log. Anything else goes to a person.
Acts, with approval for what matters
Anything irreversible or high-value is prepared by the agent and waits for human approval. Corrections feed the test set.
What an AI agent
costs in the UK.
Indicative Magora pricing for 2026, in GBP. Every engagement starts with a free working prototype, and the final figure is fixed after Discovery, when the workflow, integrations and data are known. Adding AI to an existing application typically runs £20,000–£80,000; ongoing support is from £1,000 a month.
Working prototype
A coded prototype of your agent on your own example documents or data, before you commit to anything.
Discovery sprint
Workflow map, baseline, risks, data-protection review and a fixed estimate — yours to keep, whoever builds it.
AI proof of concept
The agent running on real data in a controlled pilot, with evaluations that show whether it is good enough to scale.
AI agent for internal automation
One production workflow automated end to end — document, process or triage agent — with guardrails, audit log and monitoring.
Knowledge agent (RAG)
Retrieval pipeline, vector database and API over your document estate, answering with citations and access control.
Production LLM application
Customer-facing or multi-agent systems at production grade. Larger multi-agent programmes typically take 3–6 months.
How the price becomes fixed: fixed-price software development UK. Compare ranges across project types on UK software development pricing, estimate your own with the app cost calculator, or see the wider AI and ML development services.
From workflow
to working agent.
A production AI agent for a single workflow typically ships in 6–8 weeks after Discovery.
Workflow audit
1 weekWe sit with the people who do the work, map every step and exception, and record how long it takes today. Some workflows fail this test: a simple automation or an off-the-shelf tool would do, and we say so.
Free prototype
5 – 10 working daysA coded agent running on a sample of your real inputs, so the first conversation about accuracy is about evidence rather than promises.
Evaluations & guardrails
1 – 2 weeksA test set of real cases with known right answers, run automatically on every change. Permissions, approval steps and data-protection rules are agreed and written down.
Build & integrate
4 – 8 weeksThe agent is connected to your CRM, ERP, inbox, document store or Teams, with every tool call logged and every failure alertable. Weekly demos on your data.
Supervised pilot
2 – 3 weeksThe agent works alongside your team, who approve or correct its output. Corrections feed the test set; results are measured against the baseline from step one.
Run & improve
OngoingMonitoring of accuracy, cost per task and drift; a runbook for your team; and a plan for when the process, the data or the model changes.
The stack
behind our agents.
Chosen per project for accuracy, latency, cost and where your data is allowed to live.
- OpenAI GPT models
- Anthropic Claude
- Google Gemini
- Llama, Mistral, Qwen
- LangChain
- LlamaIndex
- OpenAI Assistants
- Bespoke orchestration
- Tool / function calling
- pgvector
- Pinecone
- Weaviate
- Qdrant
- Hybrid search & re-ranking
- Microsoft Teams
- CRM & ERP APIs
- Email & document stores
- Custom API integration
- AWS Bedrock
- GCP Vertex AI
- vLLM / Ollama self-hosting
- UK / EU regions
- Evaluations in CI
- Guardrails & approvals
- Audit logging
- GDPR by default
- EU AI Act readiness review
AI we've
put into production.
Three projects where AI does real work inside a client's operation: live call guidance, clinical treatment plans and private data classification. Each card gives the problem, what we built and how it is used.
Empath AI
- Problem
- Call-centre agents had no way to pick up undeclared signs of caller vulnerability, such as anxiety or cognitive decline, during a live call.
- Built
- A SaaS platform that analyses vocal biomarkers in live calls, turns them into guidance prompts shown to the agent inside Microsoft Teams, and adds call management reporting and anonymous diagnosis reporting.
- Outcome
- Agents get live, empathetic guidance in the tool they already use, and managers see caller and agent-wellbeing trends without exposing individuals.
BodySync
- Problem
- Clinicians spent long hours drafting personalised treatment plans by hand and looking up exercises.
- Built
- A web app that takes structured input (diagnosis, assessment findings, treatment preferences, goals) and drafts a phase-appropriate plan with OpenAI, restricted to pre-defined peer-reviewed clinical sources. Built for HIPAA and GDPR, with one-click send or PDF export.
- Outcome
- The MVP is in use in UK clinics, and further development is planned.
University of Essex
- Problem
- Sensitive files had to be classified by confidentiality level, and cloud AI services were not allowed.
- Built
- A self-contained AI Orchestrator module running two local models: an LLM (Mistral 7B) that proposes new semantic tags and confidentiality rules for human review, and a fine-tuned entity-recognition model that applies the approved rules and writes a standard YAML file.
- Outcome
- Classification is automated without sending data to third-party cloud APIs, and the model is retrained as experts review new tags.
Also: Rollio, KMA 9x9, Interprefy. All projects →
What clients
say.
Verified reviews from our Clutch profile — 75+ verified client reviews.
“We were very happy with the overall experience.”
“Partnering with Magora for the development of the messenger was a decision we will never regret.”
“They're accessible, and it doesn't feel like I'm working with a faceless team.”
AI agents:
the honest questions.
Before you commit, read our checklist is your business ready for an internal AI agent?
At Magora, an AI agent that automates one internal workflow starts from £20,000. An AI proof of concept starts from £15,000 (4–6 weeks), a knowledge agent with a RAG pipeline from £30,000, and a production LLM application from £40,000. Adding AI to an existing application typically costs £20,000–£80,000.
Every project starts with a free working prototype and a Discovery sprint from £3,000, after which the price is fixed.
A chatbot answers questions. An AI agent takes actions: it plans the steps, calls your systems through their APIs — CRM, ERP, email, document store — checks what came back and either finishes the task or hands it to a person. Many projects start as a chatbot and become an agent once the answers are trusted.
A coded prototype takes 5–10 working days and is free. A production agent for a single workflow typically ships in 6–8 weeks. Multi-agent systems and agents spanning several departments run 3–6 months end to end.
High-volume, rules-heavy work that people do by reading and re-typing: document intake and checking, data entry between systems, first-line triage of emails and tickets, and internal questions answered from policies and manuals. Work that needs judgement on every case, or where one mistake is very costly, comes later — with a person approving each decision.
Design, not hope. The agent gets only the permissions its task needs, anything irreversible or high-value waits for human approval, every action is written to an audit log, and an evaluation suite of real cases runs on every change so a regression is caught before release rather than by a customer.
If the job is drafting and summarising for one person at a time, an off-the-shelf assistant is usually enough, and we will tell you so. A custom agent pays off when the work runs across your own systems, has to follow your rules, needs an audit trail, or has to run without someone typing a prompt.
OpenAI, Anthropic Claude and Google Gemini, or open-source models such as Llama, Mistral and Qwen when data must stay on your own infrastructure. Deployments default to UK or EU regions on AWS or GCP, GDPR compliance is built in, and an EU AI Act readiness review is part of Discovery.
You do. Source code, prompts, evaluation sets and IP transfer to you on payment, and the agent runs in your cloud account. Because it is model-agnostic, you are not locked to one AI vendor or to us. We sign an NDA on day one.
Yes, if it has a repeated, high-volume task that people do by reading and re-typing. The working prototype is free and a Discovery sprint starts from £3,000, after which you get a fixed estimate that is yours to keep. If an off-the-shelf tool does the job, we will tell you so. See our guide for small business automation.
It is a known risk for any agent that reads content it does not control, often called prompt injection. We design so that a mistake does little damage: the agent gets only the permissions its task needs, anything irreversible or high-value waits for human approval, every action is written to an audit log, and the evaluation suite is run on every change.
Rule-based automation follows steps written in advance and stops when the input varies. An agent decides the next step itself, handles variation and hands over to a person when it should. The two are often combined, and if fixed rules are all the workflow needs we will say so. See the comparison above.
With a test set of real cases with known right answers, run automatically on every change, and then a supervised pilot of two to three weeks in which your team approves or corrects the output. Corrections feed the test set and results are measured against the baseline taken before the build.
Nick is Consulting Director at Magora and runs first calls and Discovery with new clients. He reviews the advice and recommendations on this page.
Have a workflow
an agent could run?
30-minute call with a senior AI engineer. Bring one process your team repeats every day — we'll tell you whether an agent is worth building, and what it would cost.