Processing a batch of appeal documents
Input Scanned documents and attachments taken in together
Chain OCR → address normalisation → database lookup → relevant rules
Output A draft review report with the evidence and the points to check
Enterprise Multi-Agent Platform
AgentQ
Specialist agents handle documents, speech and data, passing results on to a reviewable document and plan.
Overview
One request, passed between specialists, finished as a deliverable
Ask through the portal and the right agents work through documents, knowledge and data in sequence — a workflow you can run again and again.
A hub of specialist agentsKnowledge search, OCR, database queries, analysis, translation, reports, policy review — each in one place
Orchestration across agentsEach step’s result passes to the next agent, carrying multi-stage work through to the end
Evidence and security throughoutDRM and OCR preparation, cited sources, permissions and classification, and a final check — all part of the flow
Something you can reviewNot just an answer — reports, minutes, review notes and analyses you actually use
Features
The right specialist for each piece of work
Knowledge search, drafting, transcription, policy review, data queries and secure chat — on one platform.
RAG Knowledge Search
Knowledge search with evidence
Search internal documents and policies in plain language, and see the evidence and sources behind the answer.
- Search by meaning Finds documents and clauses that mean the same thing, however they are worded
- Evidence shown The answer comes with where it came from and what else is relevant
- Scope respected Only material within your remit and clearance is searched
Standard Report
Reports in your own format
Give it the essentials and it drafts the report in your organisation’s format, ready to review.
- Choosing the format Standard templates for weekly results, research notes, market summaries and more
- Filled in automatically Results, dates and plans placed under the right headings
- Reviewing the draft Check it, edit it, then use it as the official document
Meeting Intelligence
Minutes from the recording
Upload the recording and the handouts; it separates who said what and writes up the discussion and the follow-up.
- Speech recognition Transcription with speaker separation turns each contribution into text
- Summary Agenda, decisions, owners and dates, sorted out
- Finished minutes Written into your standard minutes template, with the attachments taken into account
Document Review
Checking documents against your rules
Drafts and reports are compared against internal rules and guidance, flagging possible breaches and what to shore up.
- Preparing the document PDF, DOCX, HWP and scans converted into something readable
- Compared with the rules Relevant clauses found in the chosen rules and guidance, and compared
- Findings sorted Possible breaches and gaps, ordered by severity and by the evidence for each
Text2SQL Analytics
Querying data in plain language
No SQL needed — ask the question and it fetches the data and shows it as tables and charts.
- Reading the question What to look at and under what conditions, worked out and turned into SQL
- Within your permissions Only the databases and fields you are allowed to see
- Results made visible The result and the conditions behind it, as a table, a chart and a short explanation
Secure LLM Chat
Secure chat on a local model
Sensitive questions and documents never leave the building — everything runs on a local model inside your network.
- Nothing kept Each session stands alone; questions and answers are not stored on the server
- No outbound connection All reasoning and document handling happens on the internal model
- Never used for training What you type and what it reads are never used as training data
Runtime Pipeline
How raw material becomes something you can review
Documents and data are put in order first; then evidence, reasoning and checking; then the document, the approval and the handover.
STEP 1
Intake
Text, speech, documents, spreadsheets and data queries, all accepted
STEP 2
DRM and OCR preparation
Protected files unlocked and scans read, leaving something searchable
STEP 3
Retrieval
Evidence and sources found across documents, rules and the vector database you are permitted to see
STEP 4
Reasoning on the in-house model
The internal model summarises and analyses for the task, then hands on to the next agent
STEP 5
Checking and security
Whether the evidence matches, how confident it is, personal data and classification — checked, with fixes requested
STEP 6
Document, approval, handover
The official document and the analysis are produced, reviewed, approved and sent on to the business system
Results passed on automaticallyOCR, lookups and analysis from the previous step carry forward — nothing is typed in twice
Permissions held throughoutWhat you may see, and how each document is classified, applies from the first search to the final output
A method to match the material Unstructured documents, structured data and multi-step work each go to the tool and agent that suit them.
Documents and rules · RAG
Finds and combines the closest evidence across internal documents and rules, and shows the source and its classificationalongside it
Tables and figures · Text2SQL
Turns the question into a database query, does the sums, and explains the result with tables, charts and the conditions used
Multi-step work · Multi-agent
OCR, address normalisation, database lookup, analysis and reporting joined up into a single deliverable
Prototype Workflows
Each agent takes its part, and the result comes together
The prototype shows separate capabilities chained in order, producing work a person can then review.
An unusual-transaction review
Input The transaction conditions and the data to examine are specified
Chain Database lookup → analysis → internal rules and legislation → report
Output A review document setting out the warning signs, the reasoning and what to check next
Ideal Use Cases
Where it makes the most difference
For work where documents, data and rules are tangled together, and the output needs both evidence and control.
Lots of documents, no two alike
Protected files, scans and recordings that must be put in order by text and speech recognitionbefore anything can be searched, summarised or reviewed
Work that must cite the rule it rests on
Finding the internal rule, the law or the guidance, recording the source and comparing it with the document — pre-review and assessment work
Where the same queries come round again and again
Querying databases and spreadsheets in plain language, then producing statistics, outlier checks and chartson top
Multi-step work someone currently retypes
Text recognition, address normalisation, data lookup, rule search and report writing, joined into one workflow across several agents
Where the output has to be an official document
Analysis and evidence written into the set format — report, minutes, review note — and put through approval by the person responsible
Where nothing may leave, and everything must be controlled
Running an in-house small language model and knowledge base, with permissions, classification and usage records held to one consistent standard