See
exactly
how
work
really
happens.
Understand your team's productivity in real time. Turn that same signal into consent-gated training data for AI that learns from real work. One lightweight sensor does both.
One platform.
Two SUPERPOWERS.
Whether you want a clear picture of how your team works, or a legitimate way to train AI on real work, MARFI runs both from the same sensor.
Know how work actually gets done.
Replace guesswork and self-reported timesheets with a live, honest picture of where time goes across your organization.
- Time per app, site, and category for every sensor
- Per-device drill-downs with live system telemetry
- Productivity classification and idle detection built in
- Recurring workflows surfaced automatically from real data
Train agents on real workflows
Build a proprietary dataset of how your team actually uses software, captured with explicit consent and converted to JSONL training data.
- State, action, state trajectories with before and after keyframes
- One-click JSONL export straight into your training pipeline
- Every capture tier is opt-in, per person, with a full audit trail
- Screens, actions, and system state aligned on one timeline
One sensor.
The whole picture.
Installs in 30 seconds. Then everything a modern operations and AI team needs to understand real work and train on it, all without compromising the people doing it.
Real-time visibility. Every sensor, LIVE.
Live recording status, ingestion rates, and per-sensor drill-downs across every enrolled machine, refreshed continuously, so the picture you're looking at is the one happening right now.
Consent-gated capture tiers
Activity, screenshots Trajectory, actions, and system metrics are separate opt-in tiers. Each sensor shares only what its user agreed to.
Deep system metrics
CPU, memory, GPU, disk, power, displays, and network snapshots attached to every capture. Full state for analytics and visual replay Trajectory.
Enterprise tenant isolation
Built by security experts, to include row-level security, tenant isolation, at-rest and in-transit encryption. We keep every organization's data strictly, provably, separate.
Workflow mining
Recurring task sequences surface automatically from session data, revealing the real processes behind the org chart.
AI-ready trajectories, EXPORT-READY. Trajectory
State → action → state sequences with before/after keyframes, one click away from JSONL and straight into your training pipeline. Proprietary agent-training data that grows every working hour.
Sensor to AI-ready data.
Setup in 5 MINUTES.
Enrollment is a one-time token. Everything after that is automatic, and consent-gated.
Link sensors
Mint a scoped one-time provisioning token from the console and enroll any macOS, Windows, or Linux machine in seconds.
Capture with consent
Each user opts into capture tiers. Sensors stream activity, actions, and system metrics; screenshots on Trajectory, consent-gated, at up to 1-second cadence.
Train & analyze
Watch work unfold live, mine workflows, and export state-action trajectories as JSONL to feed your models.
See where every hour REALLY goes.
Drop your work email and we'll spin up your workspace. We look up the rest, then email you a sign-in link.
Verify your company, add a payment method, deploy your first sensor — minutes, not meetings.
- Workspace ready in minutes, not a sales call
- Self-hosted or managed deployment options
- Privacy & consent model built in
Get started
No spam, no drip campaigns. A human is in the loop.
Two missions. One SENSOR.
Per-sensor pricing that matches the two superpowers: run Insight to see how work happens, or Trajectory to train on it. Retention included. No meters, no token fees, no quote requests.
The live, honest picture of how work happens. Billed upfront, prorated to the day; sensor removals take effect at renewal.
- Full live console: fleet, apps, sites, activity timelines
- Productivity analytics and workflow mining
- System metrics and category breakdowns
- 90-day rolling retention, included
- AI assistant included, no token fees
Expert demonstration data sells for $50+/hour elsewhere. This is every working hour — real, consented, export-ready.
Everything in Insight, plus the AI-training pipeline built on your team's real work.
- Screenshot capture & visual replay (consent-gated)
- State-action trajectory assembly
- One-click JSONL export in provider fine-tune formats
- Hosted fine-tune kickoff (OpenAI)
- 12-month retention, included
Your infrastructure, your walls, no data ever leaves. The full platform, deployed inside your perimeter. CMMC & NIST aligned.
- Runs entirely on infrastructure you control
- Annual license, 50-sensor minimum
- White-glove deployment and support
Your work. Your data. FULL STOP.
You own 100% of your raw captures and every derived trajectory. MARFI takes no license to your data and trains no models on it. The trajectories your team build are yours alone. You may integrate your own LLM key and all data is encrypted at-rest and in-transit.
Your key. Your provider. NO FEE.
Bring your own LLM key. Encrypted at rest (AES-256-GCM), decrypted only in-memory per request, never logged. You pay your provider directly. MARFI adds no fee.
Seats bill upfront, prorated by days of service when a sensor is added mid-cycle; removals apply at the next renewal. Retention is included: 90 days on Insight, 12 months on Trajectory. Storage usage and retention window are always visible in the console. Enclave is licensed annually and deployed on your infrastructure.
Straight answers.
No sales fog.
01What counts as a sensor?
One enrolled machine (macOS, Windows, or Linux). Billing is per sensor, upfront, prorated to the day you enroll it; removals take effect at your next renewal.
02My people have two computers. Do I pay twice?
A sensor covers a machine, not a person. Most teams deploy on the machine where the real work happens. If you genuinely need both captured, that's two sensors, and on Trajectory, two trajectory streams.
03Does Insight capture screenshots?
No. Insight is telemetry-only: apps, activity, actions, and system metrics. Screenshot capture and visual replay are Trajectory features, and even then, each user must individually opt in. The server rejects screenshot data from Insight workspaces outright.
04Can employees see what's captured?
Capture is consent-tiered per user: activity, screenshots (Trajectory), actions, and system metrics are separate opt-ins with a full audit trail. Nothing is captured in a tier the user didn't grant.
05Who owns the data?
You do: raw captures and every derived trajectory, 100%. MARFI takes no license and trains no models on your data.
06How long is data kept?
90 days rolling on Insight, 12 months on Trajectory, enforced automatically. Expired data and its stored objects are deleted, and deletion failures are surfaced, not swallowed. Custom windows available on request.
07How is data secured?
Per-tenant isolation with row-level security in the database, AES-256 encryption at rest in object storage, TLS in transit.
08How does the AI assistant bill?
It doesn't. Bring your own LLM key (OpenAI, Anthropic, Google, xAI), encrypted at rest with AES-256-GCM, decrypted only in-memory per request, never logged. You pay your provider directly; MARFI adds no fee.
09What do I need before deploying sensors?
A company email domain (free-mail providers can't create workspaces), multi-factor authentication, a confirmed company profile, and a payment method on file. Then enrollment tokens are one-time and scoped.
10Can I upgrade or downgrade?
Yes, self-serve. Upgrades to Trajectory take effect immediately (prorated); downgrades apply at renewal. Downgrading stops new screenshot capture; already-stored data simply ages out under retention.
11What's Enclave?
The full platform, self-hosted on your infrastructure, no data ever leaves your perimeter. Annual license, 50-sensor minimum. Request it from inside your workspace; a human replies.
12What format is the training export?
State→action→state trajectories with before/after keyframes, streamed as JSONL in provider fine-tune formats. Nothing is stored server-side during export. OpenAI hosted fine-tune kickoff is one click; other providers get clean generic JSONL.
Ready to see
the whole picture?
Deploy your first sensor in minutes. Watch the whole picture assemble itself, live.