AonHiveSoftwareStart a project
Navigation

Have a complex product or platform problem?

Book a 30-min callhello@aonhive.com

Applied AI

RAG · Copilots · Voice · Agents · Applied ML

An AI development company.

What can AonHive build with AI?

Knowledge assistants, copilots, voice, documents, and agents. We test them on your data, then put them into the business.

Discuss an AI projectSee our work
Letter from the OpenAI Partner Network congratulating the AonHive team on reaching OpenAI Select Partner status.

OpenAI Partner Network

OpenAI Select Partner.

AonHive is an OpenAI Select Partner. As part of the OpenAI Partner Network, we help organizations build, deploy, and scale AI solutions with OpenAI.

What we build

Find the right starting point.

From answers in your company knowledge to tools that act on them. We build around the problem your team needs to solve.

  1. 01

    RAG and company knowledge

    Turn policies, manuals, and internal documents into answers with source references and access controls.

  2. 02

    AI copilots and task agents

    Help staff prepare quotes, research accounts, and complete tasks across tools, with approval for important actions.

  3. 03

    Voice and conversational AI

    Handle enquiries, qualify leads, and book appointments over voice or chat, with a clear handoff to your team.

  4. 04

    Document and visual understanding

    Extract useful fields from PDFs, scans, and images. Validate the results and send uncertain cases for review.

  5. 05

    AI analytics and forecasting

    Ask questions of your business data, spot unusual activity, and test forecasts against historical results.

  6. 06

    Recommendations and personalization

    Bring relevant suggestions into your product, using customer context and feedback to improve their usefulness.

  7. 07

    MCP and connected AI tools

    Let assistants read records and use approved tools in your existing systems, with permissions and an audit trail.

  8. 08

    Private LLM deployment

    Run DeepSeek, Kimi, Qwen, and other open models on servers you own, so customer data never leaves your network.

  9. 09

    Model adaptation and AI operations

    Choose and adapt models using real examples. Track answer quality, response time, and running costs.

RAG, explained

Let people ask questions.
Let your knowledge answer.

Retrieval-augmented generation (RAG) searches approved business information before generating an answer. We build the source connections, retrieval, permissions, and tests behind that experience.

Illustrative source

Support handbook · § 3.2

Damaged deliveries

Record the order number and request photographs of the packaging and the damaged item.

[1] Source passage

The source is retrieved only if this user has access. The answer keeps its reference.

Illustrative answer

Team member

What should I ask for when a delivery arrives damaged?

Knowledge assistant

Ask for the order number, a photograph of the packaging, and a photograph of the damaged item. [1]

[1] Support handbook · § 3.2
  1. 01

    Connect

    Approved documents, wikis, product information, and records.

  2. 02

    Prepare

    Split and index the content, retain references, and keep it updated.

  3. 03

    Retrieve

    Find relevant passages and filter them to what this user may access.

  4. 04

    Answer

    Use the retrieved evidence, cite it, and handle missing information.

Example content only. Architecture reference: Microsoft’s RAG guide

Workflow examples

See automation at work in four business situations.

Practical examples of agents, retrieval, and rules working together. Follow the incoming request, the AI work, and the point where a person takes over.

Follow the connections from input to result.

01 Sales & outreach

Lead qualification and sales handoff

A promising enquiry sits in the ERP while your sales team works through the queue.

From incoming work to a useful resultIllustrative workflow
  1. 01

    Business trigger

    An enquiry arrives

    A form, ad, or import creates a lead in your ERP.

    Need
    A system for 12 sales reps
    Timing
    This quarter
  2. 02

    AI does the work

    AI starts the conversation

    Outreach uses the lead record, then captures intent and requirements.

    Channel
    Call or chat
    Capture
    Need · budget · timeline
  3. 03

    Rules & judgment

    Check the fit

    Qualification rules decide the next step. The agent does not invent a deal.

    Qualified
    Book a meeting
    Not ready
    Nurture sequence
  4. 04

    Business result

    A prepared handoff

    The sales rep receives the meeting and context for a useful conversation.

    Calendar
    Meeting booked
    ERP
    Notes attached
Where a person steps in

Your sales rep owns the close. Qualification rules and outreach boundaries are agreed before launch.

What changes for your team

Sales starts with a qualified conversation, rather than a cold record.

02 Orders & operations

ERP sales order automation

An order arrives as a PDF. Someone has to copy every line into the ERP and check it by hand.

From incoming work to a useful resultIllustrative workflow
  1. 01

    Business trigger

    An order lands

    An email attachment or portal PDF enters the order workflow.

    Document
    Customer purchase order
    Line item
    CBL-24 × 20 units
  2. 02

    AI does the work

    Read the document

    AI extracts structured fields from the customer's format.

    Extract
    SKU · quantity · dates
    Source
    Original document retained
  3. 03

    Rules & judgment

    Validate in the ERP

    Stock and pricing are checked against your system of record.

    Pass
    Create sales order
    Mismatch
    Route for review
  4. 04

    Business result

    The order is ready

    A valid order is created and the customer receives confirmation.

    ERP
    Sales order created
    Customer
    Confirmation sent
Where a person steps in

Missing SKUs, stock shortages, and price mismatches go to an operations review queue.

What changes for your team

Operations and finance work from one order record, without re-keying the PDF.

03 Customer support

Support triage and resolution

A customer is waiting while a support agent searches several systems for the right answer.

From incoming work to a useful resultIllustrative workflow
  1. 01

    Business trigger

    A customer asks

    Email, chat, or your help desk receives a new support request.

    Question
    Where is my replacement?
    Context
    Existing ticket and order
  2. 02

    AI does the work

    Find the evidence

    AI classifies the issue and retrieves approved documentation.

    Retrieve
    Relevant policy passages
    Reference
    Source citations
  3. 03

    Rules & judgment

    Draft for a person

    A suggested response reaches the support agent with its evidence.

    Agent
    Approve · edit · reject
    Customer
    Nothing sent yet
  4. 04

    Business result

    Reply and learn

    The approved reply is sent. Resolution is logged for evaluation.

    Reply
    Approved response
    Review
    Did it solve the issue?
Where a person steps in

A support agent approves the answer before it reaches the customer.

What changes for your team

Less searching for the team, with a person still accountable for the response.

04 Finance & documents

Invoice and document processing

Finance opens every invoice to compare its fields with a purchase order and goods receipt.

From incoming work to a useful resultIllustrative workflow
  1. 01

    Business trigger

    An invoice arrives

    A supplier PDF, scan, or EDI file enters the finance workflow.

    Document
    Supplier invoice
    References
    PO and goods receipt
  2. 02

    AI does the work

    Extract the fields

    AI reads the vendor, line items, tax, and amounts into a structured record.

    Vendor
    Supplier identified
    Fields
    Lines · tax · total
  3. 03

    Rules & judgment

    Match the records

    Business rules compare the invoice with purchasing and receipt data.

    Match
    Ready to post
    Discrepancy
    Named reviewer
  4. 04

    Business result

    Post with a trail

    Matched invoices post to accounting. Exceptions retain the approval history.

    Accounting
    Invoice posted
    Audit
    Decision recorded
Where a person steps in

Missing POs, duplicates, and amount differences pause for a finance reviewer.

What changes for your team

Routine matching moves through; the finance team spends its attention on exceptions.

Model Context Protocol

Power your existing system with MCP.

The Model Context Protocol lets Claude, ChatGPT, Gemini, and other MCP clients read and act on live business data — without replacing the ERP system or databases you already operate. We design and build the MCP server.

ERP systemDocs & APIsCustom MCP serverClaudeChatGPTGeminiCursor & Copilot
  • ERP system

    → Custom MCP server

  • Docs & APIs

    → Custom MCP server

  • Custom MCP server

    → Claude · → ChatGPT · → Gemini · → Cursor & Copilot

  • Claude

    Workflow output

  • ChatGPT

    Workflow output

  • Gemini

    Workflow output

  • Cursor & Copilot

    Workflow output

  1. 01

    Ask the chat you already use

    Live orders, inventory, tickets, and docs in Claude, ChatGPT, or Gemini — instead of exporting CSVs and pasting them into a thread.

  2. 02

    Take actions with a checkpoint

    Create a record, update a ticket, or trigger a workflow. High-stakes calls still pause for a human before anything commits.

  3. 03

    Build once, connect many clients

    One MCP server for Claude Desktop and chat, ChatGPT custom connectors over HTTPS, Gemini CLI, Cursor, GitHub Copilot, and VS Code.

  4. 04

    Keep auth and audit

    Scoped tools, credentials that never live in the prompt, and a trail of what the model was allowed to do.

Works with the clients your team already uses

  • Claude
  • ChatGPT
  • Google Gemini
  • Cursor
  • GitHub Copilot
Discuss MCP for your stackRead the MCP guide

How AonHive helps

From a useful idea to something your team can run.

Bring the problem, the data you have, and the systems involved. We help define the first build and what success should look like.

  1. 01

    Choose a useful starting point

    We review the business need, available data, integrations, and operating constraints. The first deliverable is a scoped use case and a way to judge whether it works.

  2. 02

    Prove it on your examples

    We build a bounded prototype against representative questions, documents, or tasks. Your team reviews the outputs before we commit to a production approach.

  3. 03

    Build, connect, and hand over

    We deliver the interface, data connections, permissions, evaluations, and monitoring, with documentation and ownership for the team that will run it.

Technology stack

Models, orchestration, and data — chosen for the workflow.

Explore all technologies
  • Node.js
  • Supabase
  • PostgreSQL
  • Redis
  • pgvector
  • Pinecone
  • OpenAI
  • Anthropic Claude
  • n8n
  • LangChain

Common questions

Working with AonHive on AI.

01Do you build chatbots or production AI systems?

Both — but we start with the workflow, not the interface. A chatbot is one surface. We also build document routers, classification pipelines, agent orchestration, and automation behind existing products.

02How do you handle AI failures in production?

Every system we ship has a defined failure path: fallback responses, human escalation, logging, and evaluation loops. We design for the case where the model is wrong — because it will be, sometimes.

03Can you work with our existing ERP system?

Yes. Most AI work connects to systems you already run — Salesforce, custom ERPs, ticketing tools, accounting platforms. We scope integrations as part of the workflow design.

04How long does a first AI integration take?

A feasibility spike can run in two to four weeks. A production integration depends on data readiness, integration complexity, and evaluation requirements — we scope the first milestone in the initial discussion.

05What is MCP (Model Context Protocol)?

MCP is an open standard that lets AI clients call tools and read data from your systems through a single server. Instead of building a separate integration for every chat product, you expose scoped tools once — query orders, search docs, create tickets — and connect the clients your team already uses.

06Can we use MCP in Claude, ChatGPT, and Gemini?

Yes. Claude Desktop and Claude chat connect to local or remote MCP servers. ChatGPT custom connectors use remote HTTPS endpoints. Gemini CLI and many developer tools — Cursor, GitHub Copilot, VS Code — also speak MCP. We build the server; you connect the clients you already pay for.

Applied AI

Bring the business problem you want to solve.

Tell us who needs help, what data you have, and what a useful result would look like. We'll help scope a starting point, test the approach, and plan the build.

Discuss an AI projectBook an AI consult