indentia.ai

Data integration and decision intelligence

The sovereign context layer for decision-grade AI.

Indentia fuses your systems into one governed knowledge graph, so agents answer with citations, inside your perimeter.

100% EU data residency Self-hosted & air-gapped GDPR by design
Built for Defence Government Banking & Insurance Healthcare R&D

The platform

One coherent platform, from raw data to governed agents.

Most AI stacks are loosely tied together: a vector database here, a search engine there, a policy tool next to it, an agent framework on top. Indentia is designed from the ground up as one whole. Your tenancy, identities and compliance live in the foundation, not bolted on as the last sticker.

Your data

Structured
Databases, ERP, CRM, tables
Unstructured
Documents, email, chat, video
IoT
Sensors, telemetry, edge devices
indentia
Cognitive layer
Your organization's Second Brain

Ingest · ontology · permissions · AI agents · audit, in one coherent platform.

Available as

Searchable
Hybrid search with citations
Agent-ready
Multi-step agents with scoped permissions
Deep analysis
Trends, anomalies, cross-source insights

Measured performance on MultiHop-RAG.

The public MultiHop-RAG benchmark tests retrieval and reasoning across documents. It contains 2,556 questions based on 609 news articles. Explore the earlier publication and the latest measured retrieval results.

Measurement date
Evidence recall@10
98.24%
MRR@10
0.9832

Complete tuning run: 213 questions, including 180 with reference evidence. The chart also shows the broader preliminary validation.

Scroll the chart sideways to see all metrics.

Indentia · MultiHop-RAG retrieval Indentia · MultiHop-RAG retrieval Finding and ranking evidence · score from 0 to 1 Tuning (n=180) Validation (n=498) 0 0.2 0.4 0.6 0.8 1 Score 0.98240.983210.95 Tuning (n=180) · Evidence@10: 0.9824074074074075 Validation (n=498) · Evidence@10: 0.961178045515395 Tuning (n=180) · MRR@10: 0.9832010582010582 Validation (n=498) · MRR@10: 0.9774096385542169 Tuning (n=180) · Hits@10: 1 Validation (n=498) · Hits@10: 0.9939759036144579 Tuning (n=180) · All evidence@10: 0.95 Validation (n=498) · All evidence@10: 0.8955823293172691 Evidence@10 MRR@10 Hits@10 All evidence@10 Metric

Broader validation, preliminary: 96.12% Evidence recall · 0.9774 MRR. 569 of 2,343 questions processed, including 498 with reference evidence.

Earlier publication from 19 May 2026. Answer quality and retrieval on the previously reported set of 2,255 questions.

Scroll the chart sideways to see all metrics.

Indentia v5 · MultiHop-RAG (n=2,255) Indentia v5 · MultiHop-RAG (n=2,255) May 2026 publication · answer quality and retrieval Paper (GPT-4) World best Indentia v5 0 0.2 0.4 0.6 0.8 1 Score 0.960.920.690.620.720.640.980.69 Paper (GPT-4) · Inf F1: 0.56 World best · Inf F1: 0.95 Indentia v5 · Inf F1: 0.96 Paper (GPT-4) · Inf Acc: 0.56 World best · Inf Acc: 0.95 Indentia v5 · Inf Acc: 0.92 Paper (GPT-4) · Comp F1: 0.56 World best · Comp F1: 0.67 Indentia v5 · Comp F1: 0.69 Paper (GPT-4) · Comp Acc: 0.56 World best · Comp Acc: 0.67 Indentia v5 · Comp Acc: 0.62 Paper (GPT-4) · Temp F1: 0.56 World best · Temp F1: 0.585 Indentia v5 · Temp F1: 0.72 Paper (GPT-4) · Temp Acc: 0.56 World best · Temp Acc: 0.585 Indentia v5 · Temp Acc: 0.64 Paper (GPT-4) · Hits@10: 0.7 World best · Hits@10: 0.74 Indentia v5 · Hits@10: 0.98 Paper (GPT-4) · MRR@10: 0.435 World best · MRR@10: 0.435 Indentia v5 · MRR@10: 0.69 Inf F1 Inf Acc Comp F1 Comp Acc Temp F1 Temp Acc Hits@10 MRR@10 Metric
The measurement dates use different question selections and metrics. They do not form a direct before-and-after comparison. MultiHop-RAGMeasurement details and whitepaper
01

Inference

Connecting facts across sources. "What can we conclude from these reports?"

02

Comparison

Comparing two or more sources. "Does view A differ from view B?"

03

Temporal

Judging time-sensitive claims. "Is this still current?"

What the latest results mean: evidence recall measures how much of the required source material reaches the first ten results. MRR measures how close the first relevant source is to the top.

The September results measure source selection and ranking. Answer accuracy is a separate measure. Read the whitepaper and measurement details.

Autonomous agents

A specialised agent for every business process.

With Indentia, you can create domain-specific agents that understand a particular role, department or workflow. One agent can review cases, another can prepare reports, while others check compliance or update Salesforce. Each agent works with your terminology, business rules and approved data sources.

These specialists can collaborate and hand work to one another. Each receives only the tools and permissions it needs, while people remain in control of sensitive decisions and every step stays traceable.

Domain-specific agent architecture: a question is routed by the Indentia coordinator to case, reporting, compliance and custom agents that share one secure business context and return a checked result.

One stable identity per thing, across every source system.

Every entity gets a universal Indentia IRI under id.indentia.ai. One customer, one contract, one project. No more ID spaghetti between Salesforce, ERP and SharePoint.

Two John Smiths? No problem. The platform tells apart people, projects and organizations with the same or similar name by their context, role and relationships, so a question about "John on case X" doesn't return the answer about "John on case Y". Off-the-shelf RAG stacks hallucinate precisely here: they collapse the two people into one blurry profile. Essential for handling ambiguous terms in legal, HR and intelligence contexts.

See how it works

indentia · iri-template.ttl
# Universal Indentia IRI template
https://id.indentia.ai/{domain}/{type}/{tenant}/{source?}/{id}

# Example: chat message in tenant `acme`
https://id.indentia.ai/messaging/ChatMessage/acme/teams/a3f9…b21

# Domain prefixes
imsg:   // messaging
idoc:   // documents
imedia: // media
iintel: // intelligence
iid:    // identity
igov:   // governance

Your data stays in Europe. No exceptions.

EU-only, fully self-hosted or even air-gapped. No US Cloud Act exposure. Permissions are inherited from source systems (least privilege), SSO integration respects existing authorizations, and every action is logged for traceability.

100%
EU data residency
5
Security layers per agent
100%
Actions audit-logged

Want to see this on your own data?

Request a demo, or start a short 2-4 week pilot on two or three sources to make time-to-value and the impact on your most important processes measurable. A pilot without a pre-agreed yardstick is a demonstration, not evidence, so we agree on the measure before we start.

Request a demo See pricing