An assistant that respects who is allowed to see what
Enterprise knowledge systems where retrieval inherits your existing permissions — because an assistant that leaks a salary review is worse than no assistant.
The demo works. Then someone asks a question they should not be able to ask.
Internal knowledge assistants are easy to prototype and hard to deploy, because the prototype indexes everything with a service account. The moment real employees use it, the index becomes a way to read documents the person was never granted.
Permissions travel with the document, all the way to the answer
We carry the source system's access control into the index itself, filter retrieval by the signed-in user's entitlements before anything reaches the model, and cite every claim back to a document the user can actually open. If they cannot open it, it cannot be retrieved.
Capabilities
Enterprise Knowledge Assistants
Question answering over internal documentation, scoped to each employee's access.
RAG Systems
Retrieval-augmented generation with citations, evaluation and a measured baseline.
Document Intelligence
Extraction and classification across contracts, policies and operational documents.
Internal AI Search
Semantic search across systems that each have their own, worse search.
AI Knowledge Bases
Curated, versioned content designed to be retrieved rather than merely stored.
AI-powered Reporting
Narrative summaries generated from governed data, with the figures traceable.
AI Copilots
Role-specific assistants embedded in the tools a team already has open.
How it fits together
Secure retrieval architecture
The secure retrieval layer is the load-bearing component. It filters by entitlement before the model sees anything — the model itself is never a security boundary.
- 01Company documentsWith their source ACLs
- 02SharePoint
- 03Policies
- 04Database
- 05Knowledge base
- 06Secure retrieval layerFilters by the user's entitlements
- 07AI assistantAnswers only from what was returned
- 08EmployeeSees citations they can open
The ingestion pipeline
Permissions are captured at ingestion and re-checked at query time, because entitlements change after a document is indexed.
- 01Document ingestionSource ACLs captured with the content
- 02ChunkingStructure-aware, not fixed-size
- 03Embeddings
- 04Vector searchFiltered by entitlement
- 05RetrievalRe-ranked, permission re-checked
- 06Context constructionBudgeted, deduplicated
- 07LLM generation
- 08CitationEvery claim linked to a source
- 09Access controlEnforced again on the citation link
- 10EvaluationAnswer quality and retrieval recall
- 11MonitoringUsage, cost, refusals, drift
Systems we build against
Platforms and services TOOKLI actively works with on client engagements.
Content sources
- SharePoint
- Microsoft 365
- Teams
- Confluence-style wikis
- File shares
Retrieval
- pgvector
- PostgreSQL
- Azure AI Search
Models
- Azure OpenAI
- OpenAI
- Anthropic
Identity
- Microsoft Entra ID
- SAML
- OIDC
This list reflects systems we integrate with, not partnerships, resale agreements or certifications. If something you rely on is missing, ask — we will tell you honestly whether we have used it.
What this looks like in practice
Examples of what we can build for each sector. Work we have actually delivered appears in our case studies, with named outcomes.
- Professional services
Answering policy and process questions
The questions that currently interrupt a senior colleague several times a day.
- Construction
Finding the clause in the specification
Search across drawings, specifications and change orders, with the source cited.
- Healthcare
Administrative document assistance
Summarising and locating administrative material. Not clinical decision support.
- Finance
Internal reporting and policy retrieval
Internal-facing only, with every generated figure traceable to a governed source.
Delivery process
- 01
Inventory the sources
What exists, who owns it, how current it is, and who is allowed to read it.
- 02
Model the permissions
The hardest part, done first. If entitlements cannot be resolved, nothing else matters.
- 03
Build the ingestion pipeline
Chunking, embedding and ACL capture, re-runnable and incremental.
- 04
Establish an evaluation set
Real questions with accepted answers, including ones the system should refuse.
- 05
Build retrieval and generation
Filtered search, re-ranking, context budgeting, mandatory citation.
- 06
Pilot with one department
Narrow scope, real users, measured. Expand only after the numbers hold.
- 07
Monitor and re-index
Content changes, permissions change, models change. The pipeline is a running system.
How we keep it safe and accountable
Engineering practices, not certifications. TOOKLI holds no security certifications and does not claim any.
Permissions are preserved end to end
Source ACLs are captured at ingestion and enforced at query time. A user cannot retrieve what they cannot already open — no exceptions, and no reliance on the prompt.
Citations, always
Every answer links to its sources. An uncited answer is treated as a failure, because an employee cannot verify what they cannot trace.
Refusal is a correct answer
When retrieval returns nothing relevant, the system says so. Confidently wrong is the expensive failure mode in an enterprise setting, and we measure it separately.
No confidential content leaves without a decision
Which content is sent to which provider is an explicit, documented choice. Where data residency requires it, we deploy against models hosted in your own tenancy.
Auditability
Who asked what, what was retrieved and what was returned — logged and reviewable, which is usually a condition of approval from security and legal.
Common questions
Related services
AI Agents & Intelligent Automation
Agents and automated workflows that connect your data, applications and processes — built with human approval, permission controls and monitoring from the first design.
Learn moreAI Integration
Bring AI capabilities into your existing applications — search, extraction, classification, summarisation and assistants — without building or training models yourself.
Learn moreAI, Data & Business Intelligence
AI-assisted dashboards, natural-language analytics, automated reporting and data quality automation, built on a semantic layer people already trust.
Learn more
Have an internal knowledge problem?
The first conversation is about permissions and sources, not about models. That is usually where the project is won or lost.