Answers are easy. Evidence is harder.

Trace lets teams explore large document collections, uncover recurring themes and see how documents connect, without losing the sources behind each finding. Perceptor prepares difficult scans, tables and technical drawings when the collection needs structure first. Both products run in the environment you control.

See Trace in action Talk to an expert

Every claim should lead somewhere.

AI-generated illustrative visual

TRACE

A conclusion is only as strong as its evidence trail.

Trace lets teams investigate large document collections, build shared contexts and reveal relationships while every finding remains tied to its sources, filters and contradictions.

Book a Trace demo

Build a shared context

Contexts are curated document spaces for recurring work. Teams can assemble, extend and investigate a body of material together.

Useful forDue diligenceIncident responseRecurring research topics

Context

Sicherheitsrichtlinie

Curated document space · shared with 4 researchers

Selected material

Richtlinie 4.228 pages · 2019
Protokollreihe42 records · 2020 to 2023
Audit-Anhänge13 files · manually added

Add source material

TimelineSources used in the current investigation

Reveal the connections across your documents

A selected theme can become a source-linked wiki and knowledge graph, revealing recurring actors, events, changes and disagreements.

Useful forStakeholder mapsPolicy lineageInstitutional knowledge

Reopen every finding

Answers retain their named sources, applied filters and contradictions. The retrieval path is available when the investigation needs it.

Useful forAudit reviewsCompliance checksFact checking

Research dossierDE

Die Richtlinie öffnete den Zugriff schrittweise, aber die Unterlagen nennen unterschiedliche Freigabezeitpunkte.

Zeitraum: 2019 bis 2024
Filter: Richtlinie · Protokoll · Audit

Ab 2021 wurden externe Teams und dokumentierte Ausnahmen aufgenommen.^[1][2] 2024 wurden Rollen und Meldewege konsolidiert.^[4]

WiderspruchProtokoll 06/21 nennt Juni als Freigabe. Der Audit-Anhang verweist auf September.^[2][3]

Ask across your documents

Start with a question in natural language. Trace keeps the investigation focused on the period, material and level of detail that matter.

Useful forPolicy historiesEditorial archivesCase files

Research questionDE

Wie hat sich die Sicherheitsrichtlinie zwischen 2019 und 2024 verändert? Wo widersprechen sich die Unterlagen?

Investigation focus

Changes over time
Conflicting records
Supporting source material

Collection: Selected internal documents
Period in question: 2019 to 2024
Answer standard: Cite every finding

Run it where your data belongs

Operate Trace, and Perceptor when needed, inside a boundary that fits the sensitivity and architecture of your organisation.

Useful forSensitive contractsRegulated recordsInternal knowledge bases

Selected operating model

Separate the boundary deliberately.

Keep the sensitive data layer local while approved compute operates in a separate controlled environment.The operating boundary remains visible.

Trace in your own collection

Bring one question. Reopen every source.

Use a real document problem and see how Trace keeps the answer, context and evidence connected.

Book your demo

PERCEPTOR

The page is a surface. The structure is the information.

Perceptor keeps the relationships inside the source intact, so a table row, a geometric reference or a critical field can move into the systems that need it.

Discuss Perceptor

AI-generated illustrative visual
Perceptor identifies a table row, a geometric reference and a named field while preserving where each value came from in the document.

One extraction interface

Documents do not stay simple. Their structure still can.

Define the document types and information you need. Perceptor reads simple fields and complex relationships across pages, layouts and languages, then returns a source-linked result through one API.

Technical drawing

Geometry, dimensions and specification tables remain related.

Multi-page document

Document type, layout, fields and language can vary between pages.

Complex table

Rows, columns and page references stay intact across extraction.

REST APIPOST /extract

200 · application/json

{
  "document_type": "technical_drawing",
  "item": "04",
  "diameter_mm": 120,
  "tolerance_mm": 0.2,
  "page": 3
}

Start immediately: Zero-shot extraction without prior training, with a path to fine-tuning or Flavor Training for maximum recognition performance without hallucinations.
Choose the workload: Real-time extraction in under 4 seconds per page, or batch processing for larger document volumes.
Integrate by API: Send a document and the information you need. Receive structured JSON with page references.
Set the boundary: Operate as a TamedAI Cloud Service, in a cloud you choose, or on-premises.

Optional connection to Trace

Structure can enter an investigation without losing its source.

Perceptor can supply Trace with structured, source-linked units. Each product also works independently.

Research notebooks and a drafting sheet on a dark worktable — AI-generated illustrative visual

Founded in 2019 · University spin-off

Built for the work behind the answer.

TamedAI was founded in 2019 to bring language models into productive information workflows, from focused systems to larger language models.

Our machine learning and software engineers build dependable products for organisations that need to extract, find, and understand information without giving up control.

As a university spin-off, we remain close to current research and contribute to projects that advance the performance and reliability of our work.

Explore careers

The people behind the work.

A multidisciplinary team working across product, machine learning, engineering and operations.Portraits generated with AI from reference photographs

Nils Schwenzfeier

CTO & Co-Founder

Dirk Müller

CEO

Andre Hönnscheidt

Machine Learning / AI Engineer

Kevin Nagel

Machine Learning / AI Engineer

Mark Günzel

Machine Learning / AI Engineer

Bergül Schulte

Administration / HR

Start with the material.

Bring a real document problem.

Tell us what your teams need to extract, find, or understand. We will show you a path that fits your environment.

Talk to an expert