Grounded by default
PropensityAI answers only when it can point to supporting evidence. Otherwise it flags uncertainty and suggests what to check next.
Evidence-first • confidence signals • refusal modes
LLM-powered search • grounded answers • citations
Find accurate answers — with sources you can trust.
PropensityAI combines retrieval + reasoning to deliver concise, cited results across web, docs, and media — built with privacy-first defaults and no training on your private queries or uploads.
Docs + Web
Retrieval + generation
Cite
Answer with sources
Multimodal
Text + image + audio + video
Enterprise AI Search
Advance Reasoning
Enterprise AI Agents
On-prem LLM Systems
Data Secure AI
250k+
Searches processed
Trusted across high-volume research and discovery workflows with grounded answers, citations, and multimodal retrieval.
Why PropensityAI
Not “chat with the internet” — a search product built for verification, speed, and control.
Citations you can follow
Every key claim maps to a source — documents, pages, transcripts, timestamps, or paper sections — so users can verify instantly.
Claim-to-source mapping • timestamps • paper sections
Multimodal coverage
Search across text, images, audio, video, and research papers in one interface — then get one clean, cited answer.
Images • audio • video • YouTube • PDFs
Privacy-first by design
Your content stays yours. PropensityAI is built to support secure deployments and does not train foundation models on your private queries or uploaded files.
No model training on user data • data minimization • retention controls
PropensityAI vs typical chat
Evidence, privacy, and control — without extra noise.
PropensityAI
- ✓ Cited answers (claim → source)
- ✓ Grounded by default + confidence signals
- ✓ Neuro-symbolic grounding (LLM + knowledge graph)
- ✓ Multimodal (text, images, audio, video, papers)
- ✓ Privacy-first: no training on private queries/uploads
Typical chat
- — Often uncited or loosely sourced
- — Mixed reliability, weak uncertainty UX
- — No knowledge graph grounding layer
- — Limited media/paper workflows
- — Unclear data usage / training assumptions
Get started
Client-side mock that shows the PropensityAI experience: answer + citations.
Type a question above to see a grounded answer with citations.
How it works
A simple, reliable pipeline: retrieve → reason → cite → guardrail.
1) Retrieve
PropensityAI pulls relevant passages from your docs, sites, or knowledge base.
BM25 + embeddings • filters • freshness
2) Reason
The model synthesizes the passages into a concise, structured answer.
Summarize • compare • compute • explain
3) Cite + Guard
Every key claim is linked to a source. Unsafe or uncertain outputs are flagged.
Citations • uncertainty • policy checks
Built for teams
Role-based access, audit trails, and source controls for enterprise workflows.
Low noise UI
Minimal interface that feels like search, but answers like an assistant.
PropensityAI Features
Multimodal search + analysis across content formats, with grounded answers.
Analyze / process images
Extract meaning from screenshots, diagrams, UI mockups, and photos — then answer with evidence.
OCR • captioning • visual QA • structured extraction
Search YouTube videos
Find the best video segments, summarize key points, and cite timestamps as sources.
Transcript search • segment ranking • timestamp citations
Analyze audios
Turn meetings, calls, or podcasts into searchable insights with summaries and action items.
Transcribe • diarize • summarize • extract tasks
Analyze videos
Search across scenes and spoken content. Pull clips, summaries, and citations by time range.
Scene detection • transcript + vision • clip references
Research papers
Ask questions across PDFs and citations. Compare methods, results, and limitations quickly.
PDF ingestion • citation graph • claims-to-source mapping
Grounded answers
PropensityAI answers only when it can point to supporting sources — otherwise it flags uncertainty.
Confidence • refusal modes • audit trails
On-prem AI appliances
Fully local, state-of-the-art LLM solutions
Own the model, own the hardware, own your data — with turnkey systems that run entirely inside your infrastructure.
No cloud dependencies. No third-party AI providers. No data leaving your walls.
Designed for companies and organizations that refuse to compromise on data privacy, these pre-configured, air-gapped systems put powerful large language models directly into your hands on dedicated hardware.
Air-gapped ready
One-time purchase
No per-token fees
Perpetual ownership
Frontier model
110B General Intelligence
from €20,000
one-time payment
A general-purpose 110-billion-parameter LLM for confidential internal assistants, sensitive document processing, and data-sovereign workflows.
80%GPQA Diamond
95%AIME 2025
89%MMLU
- Open-weight model with commercial-use license
- Pre-configured hardware and LLM bundle
- Ready for secure, local deployment
Coding specialist
30B Software Engineering
from €8,000
one-time payment
A coding-focused model for development acceleration, code review, debugging, refactoring, and complex engineering automation.
76%SWE-bench Verified
- Optimized for code generation and debugging
- Dedicated hardware plus model
- Open-weight model with commercial-use license
Reasoning engine
30B Science & Logic
from €8,500
one-time payment
A reasoning specialist for science, math, research, quantitative analysis, and rigorous problem-solving without exposing IP.
74%GPQA Diamond
- Tailored for R&D and analytical teams
- Dedicated hardware and LLM included
- Open-weight model with commercial-use license
PropensityAI Web Platform Pricing
For hosted search, grounded answers, citations, and multimodal analysis in the PropensityAI web platform.
Starter
€0/mo
For individual web search and file analysis.
- Up to 16 sources per answer
- Source Reranking option
- Multimodal: 1 file per query
- Standard + Reasoning models (best-effort; may throttle at peak)
- Basic citations
- No sign-up required
Team
Custom
For teams using PropensityAI with shared docs and connectors.
- Up to 30 sources per answer
- Multimodal: up to 10 files per query
- Standard + Reasoning models (higher availability)
- Your docs + connectors
- Access control
- Analytics + feedback
- No model training on user data
Enterprise
Custom
For secure web platform deployments, VPC, and governance.
- Up to 40 sources per answer
- Multimodal: up to 10 files per query
- Standard + Reasoning models (on-demand availability)
- On-prem / VPC
- Advanced governance
- Dedicated support
- No model training on user data
FAQs
Quick answers about PropensityAI, sources, and multimodal workflows.
PropensityAI is built like a search product: it retrieves evidence first, then generates a concise answer grounded in sources.
Citations and confidence signals make verification fast and reduce ungrounded responses.
PropensityAI uses retrieval (RAG) to pull relevant passages before generation, and it ties key claims to sources.
When evidence is weak, it can flag uncertainty or refuse instead of guessing.
Text and documents, images (screenshots/diagrams), audio (meetings/podcasts), video (scenes + transcripts),
YouTube search via transcripts and timestamps, and research papers (PDFs with citations).
Yes — typical setups include connectors to shared drives, wikis, ticketing systems, and internal docs,
with access control and audit logs so results respect permissions and governance.
PropensityAI can attach citations to sources (including timestamps or paper sections) and maintain audit trails of queries,
retrieved evidence, and generated responses for review and compliance needs.
Yes. PropensityAI can provide turnkey local LLM systems with dedicated hardware, model, and runtime environment configured to run inside your infrastructure, including air-gapped deployments where required.
No. Local LLM packages are offered as one-time purchases with no per-token usage fees. Custom fine-tuned or quantized packages can be scoped separately based on your needs.
Contact
Want PropensityAI on your knowledge base or a fully local LLM appliance? Let’s talk.
Email
contact@propensityai.com
Security
SOC2-ready roadmap • SSO