Skip to the labs
AI LabsSuper Intelligence, Research and Development
00 Overview

Introducing

RPostAI Labs

The next chapter of RPost starts here. We explore what our products can become, and then we build it.

AI Labs is where RPost researches, experiments, builds and proves new ideas, so every product keeps evolving with the technology around it.

See the three labs Jump to beta environments
RAPTOR HawkEye 3D Pre-Crime Report: a threat radar globe with red and yellow alert clusters over North America and Europe, flanked by counts of activities, senders, recipients, domains, fingerprints and countries.
Running today

RAPTOR HawkEye 3D Pre-Crime Report, Master 1.3. Real screenshot; recipients masked.

Lab index

  1. LAB 01

    RAPTOR Labs

    Threat data, intelligence, pattern detection and prediction, run as a loop so the RAPTOR ecosystem learns from itself.

    4harness layers
  2. LAB 02

    FX Studios

    Beta development environments and pitch videos, each shown as a short recording of the real thing.

    5projects in beta
  3. LAB 03

    Proprietary Super Intelligence

    Our own intelligence model, built and owned by RPost on NVIDIA hardware, not rented from a closed AI company.

    5workloads
LAB 01Active research

RAPTOR Labs

RAPTOR Labs is where our threat model gets smarter. We turn threat data into intelligence, detect the patterns behind it, predict what comes next, and run an agentic improvement loop so the whole RAPTOR ecosystem learns from itself and improves with every cycle.

  • 01Threat data
  • 02Intelligence
  • 03Pattern detection
  • 04Prediction
  • 05Agentic improvement loop

The RAPTOR Harness

One orchestrator · three agent teams · one model that learns

  • Rocky, the face of RAPTOR LAYER 1 · ORCHESTRATOR, ENGINE, MODEL

    RAPTORRuns the pipeline

    Included in every email, document and e-signature envelope. The pre-crime engine collects protocol data, the customer's policies define what is a risk, and the model analyzes openings, recipient activity, connections and devices. When the signals show risk in the content it triggers the AI Classification Agent, the AI Alert Agent that communicates to the organization, and the AI AutoLock Agent.

    OutputAlerts and locks for the organization, and the model every team reports back into.

  • OWL LAYER 2 · AT THE METADATA

    OWLEnrichment

    Sits where RAPTOR collects the metadata. Breaks the protocol data into more data points and cross-analyzes them: browser versions against OS versions, IPs, locations, the domain age of URLs. When it finds a new set of relevant data it reports back, so RAPTOR learns to identify and detect better moving forward.

    OutputNew data points, fed back into the RAPTOR model.

  • HAWK LAYER 3 · AT THE ALERTS

    HAWKCyber counterintelligence

    Sits where the alerts have been generated. Finds activity that looked normal but is suspicious, elevates the false negatives RAPTOR passed as expected, and groups patterns into threat actors, global organizations and state-backed attackers. Proves fingerprints on vast amounts of data, matches duplicates and promotes the most relevant to OSPREY.

    OutputProven fingerprints and pattern groups promoted to OSPREY: past patterns that predict future attacks.

  • OSPREY LAYER 4 · ACROSS ALL CUSTOMERS

    OSPREYGlobal intelligence

    Sits across the entire customer database, fed by RAPTOR, OWL and HAWK. Identifies widespread attacks that target different industries the same way, builds global fingerprints every industry benefits from, refines the fingerprints HAWK sends, and gives threat actors an identity and a way of operating.

    OutputNew triggers promoted to the RAPTOR model. The source of truth for the RPost proprietary model of intelligence.

Agentic Improvement Loop

RAPTOR harness · 46 s loop

The harness, assembled step by step

Click any step to jump to it. The diagram shows the order of work, not live data.

LAB 025 projects in beta

FX Studios

Working software, not slides. Every clip below is recorded from a beta development environment that runs today.

BETA 01

RAPTOR HawkEye 3D Report

A pre-crime report for CISOs and CTOs. A rotating threat radar, the trail of each email from the sender to every opening, and fingerprint matches across countries. It runs from a single file, with no server and no internet connection.

Master 1.3Standalone HTMLRecipients masked by default
  • Threat radar: globe or map view of every plotted opening
  • Review queue sorted by priority, with fingerprint profiles
  • Email trail replayed in recorded order, opening by opening
Recording of HawkEye: the dashboard loads, the High Risk Matters queue opens, a search for Russia and Nigeria finds one matter, and its email trail plays from the sender in Phoenix through the Microsoft mail server to the first opening, in Moscow. Live capture
BETA 02

RPost AI Classification

A classification platform for every RPost product. It labels content by sensitivity, from SharePoint and OneDrive or from existing Purview and Proofpoint labels, then applies the right policy wherever the content goes: RMail, RDocs, RSign and Large File Share. Rules decide first; a decision engine handles the unsure files.

Built withProspect participation: Morguard, Canada
Platform conceptRMail · RDocs · RSign · LFSMasked sample data
Recording of the AI Classification beta environment running a scan: twelve sample files are labelled one by one from Public to Highly Confidential, by rules or by the decision engine, while the counters fill. Live capture
BETA 03

RPost Product Sandbox

Interactive storytelling across platforms. A high-fidelity replica of the live products, RPost One, RAPTOR AI, RMail, RDocs and RSign, where one story moves from Outlook to the portal to RAPTOR AI, exactly as a customer would see it.

Master 1.0Outlook → RPost One → RAPTOR AI
Recording of the Sandbox story: an Outlook email with an attachment is addressed to a look-alike domain, RMail warns about the unverified recipient, and the story continues in RPost One's RAPTOR AI page. Live capture
BETA 04

RPost Interactive Product Stack Diagram

The whole RPost stack in six rings, from inbound gateway to "attack preempted", played as stories. Here, a trusted recipient forwards a message, one forwarded copy is opened from a compromised account, and RAPTOR's Active Tracker catches the unusual opening.

Master 3.94 storiesOffline single file
Recording of the Third-Party Compromise story: a web of recipients spreads out from the organization, one forwarded copy is opened, and RAPTOR flags the unusual opening with its raw transaction metadata. Live capture
FX 05
New project

Pitch video production

In-house audio and video production. FX Studios produced the RAPTOR PRE-Crime pitch videos for the Tradewinds (Department of War) and IARPA marketplaces, and a Data Sovereignty video: script, computer-generated voice-over, music, sound effects, motion graphics and animated diagram stories, rendered to 1920×1080 MP4 by RPost's own pipeline.

3 videosVoice · score · SFX · motionDiagram stories
Protocol signals fan out from your endpoints through the RPost cloud gateway to fourth and fifth parties.
TradewindsContent becomes a sensor network
Processing, open, location, network, device and agentic signals flow into the RAPTOR pre-crime model.
IARPASignals into the pre-crime model
The HawkEye globe traces one message to openings in Moscow, New York, Miami and Lagos.
HawkEyeOne message tracked across three continents
RPost AI Classification scans SharePoint and OneDrive and gives every file a sensitivity tier.
Data SovereigntyEvery file stays inside its borders
LAB 03In development

Proprietary Super Intelligence

RPost is building its own intelligence model, so we can explore new AI as it arrives and adapt quickly, instead of waiting on closed AI companies.

Our approachPowered by NVIDIA

Super-AI computeBuilt to explore what comes next

AI is moving fast, and today's best model will not be tomorrow's. Running our own model on NVIDIA accelerated computing lets RPost try new models and approaches as they appear, and adapt our products on our own schedule.

  1. TrainFine-tune RPost's open-weight model on de-identified threat patterns that OSPREY has validated.
  2. PredictScore the pre-crime risk of every opening of protected content, for RAPTOR SI.
  3. ClassifyLabel content by sensitivity for RMail, RDocs, RSign and Large File Share.
  4. MatchSearch behavioral fingerprints across proxies, tenants and countries.
  5. ExplainWrite HawkEye reports and analyst summaries, with a confidence score on every result.
  1. What it is
  2. Open weightRPost holds the model weights, so our core intelligence does not depend on a closed AI company's API.
  3. Commercially licensedA licence that allows commercial use and RPost's own tuning.
  4. Where it runs
  5. Powered by NVIDIATrained and run on NVIDIA accelerated computing that RPost operates, so the model and its improvements belong to RPost.
  6. SafeguardsOnly allowed answers, a confidence score on every result, and a person in the loop when confidence is low.
  7. What it delivers
  8. Token maximizationMore useful work from every token: short, typed outputs instead of long free text.
  9. Workflow efficiencyBuilt into the flows RPost already runs: classification, threat triage and reporting. Each OSPREY-validated fingerprint improves the next release.

RPost AI Labs

Technology will keep changing. Our research and development makes sure we change first.

RAPTOR Labsbuilds a threat model that gets smarter with every signal.
FX Studiosbuilds the next products, and the stories that launch them.
Proprietary Super Intelligencebuilds the intelligence we own, ready for whatever AI brings next.