AI agent in WhatsApp, Telegram, Slack and Teams, free · Public model cards · Unlimited guards
The first guard tour platform with anomaly detection, predictive routing, and explainable AI built in. No black boxes — every model has a public model card and a tunable confidence threshold.
AI deviation detection generally available · SOC 2 Type II, ISO 27001 and ENS certified · No cross-customer training.
AI agent in your chat apps
Talk to the PatrolTech AI agent from the chat apps your team already uses. It answers questions about rounds, scans and patrol history, and it carries out changes to checkpoints, patrollers and routes. No new panel to learn.
Any suspicious scans at Riverside Mall tonight?
One. Parking level 2 was scanned twice in under a minute at 23:05, and the round stopped there.
Create a checkpoint called Roof access door and add it to the night route after Parking level 2.
Done. Roof access door is created and now comes right after Parking level 2 on the night route at Riverside Mall.
Built in the open — every model auditable
Three production models, two in development. Every output explained. Every threshold tunable. No magic.
Isolation Forest + temporal LSTM. Flags patrols deviating from learned site patterns. 74-89% precision after 4 weeks of training data. Per-site model isolation.
Fine-tuned RoBERTa. Auto-categorizes incident reports into 47 standard categories. 91% precision out of the box. EN, ES, FR, PT, DE, IT supported in production.
Constrained optimization with learned weights. Suggests checkpoint order. Reduces average patrol time 8-14%. Production beta with 12 customer sites.
No 'the AI said so' answers in audit reports. Every output ships with the contributing factors, confidence score, comparable past cases, and a toggle to dive deeper.
Per-alert explanations show contributing factors, confidence breakdown by factor, and 3-5 comparable past cases. Audit-ready exports include the AI reasoning chain.
Every output ships with a confidence score. Tune thresholds per site (50% high recall / 75% balanced / 90% high precision). The dashboard shows precision/recall tradeoff for your settings.
Anomaly Detection trains a per-site model on your data only. Base classification models were trained on a curated public + synthetic dataset (documented in the model card). Your data never trains models for other customers.
AI suggests, humans decide. No automated decisions with operational impact. GDPR Article 22 compliant. Audit logs separate AI suggestion from human action.
Every production model has a public card with training data sources, evaluation methodology, known limitations, and bias evaluation. Read before you buy.
Inference on AWS Frankfurt (eu-central-1) for EU customers. No cross-region transfers without explicit consent. Customer-managed keys on Business plan.
We've seen the marketing. We're going to disappoint you in the right way. This list converts more procurement teams than any feature page.
We don't identify people on CCTV: privacy regulations + bias risks. We can integrate with your existing video analytics platform via webhook if needed. Facial verification and biometrics are a separate feature (beta, Business plan).
No 'this guard might be tired' alerts. Our models work on patterns, not people. Individual scoring is incompatible with our principles.
Every alert requires human review before any operational action. GDPR Article 22 by design. The AI suggests; a human decides.
Your data trains your models. Period. Base models are trained on curated public + synthetic data, documented in the model card.
Drafts, yes. Auto-submit, no. The human writes the final version, with the AI suggestion in the audit trail.
The AI agent in your chat apps answers questions and manages your patrol setup when you ask it to: checkpoints, patrollers, routes. It doesn't lock doors, page emergency services or dispatch responders. Humans do.
Built on principles, not hype. Every principle has a measurable commitment.
Every model has a public model card with training data, intended use and known limitations.
AI suggests, humans decide on any operational action. Always.
Your data trains your models, never others'. Per-tenant isolation, customer-managed keys on Business.
No 'trust me' outputs. Every prediction shows confidence and contributing factors.
Published precision/recall on every release. False-positive rates exposed in your dashboard.
Every flagged event explainable to operations and audit. No 'the model said so' in audit reports.
We update this quarterly based on public docs. Vendor claims change — we ship corrections fast.
| AI capability | guardtourai.com | TrackTik | QR-Patrol | Silvertrac |
|---|---|---|---|---|
| AI agent in messaging apps (WhatsApp, Telegram, Slack, Teams) | Yes: free plan, on by default, any language | unknown | Web app + AI clients (MCP) | unknown |
| AI assistant executes actions, not only answers | unknown | unknown | ||
| Operate the account from AI clients over MCP | Yes: Claude, ChatGPT, Gemini | unknown | unknown | |
| Anomaly detection (deviation from pattern) | ||||
| Incident auto-classification with precision shared | ||||
| Predictive routing | ||||
| Public model cards | ||||
| Confidence scores in alerts | ||||
| Customer data isolation (no cross-training) | ||||
| GDPR Article 22 compliance | ||||
| Tunable confidence thresholds |
AI agent in messaging apps (WhatsApp, Telegram, Slack, Teams)
AI assistant executes actions, not only answers
Operate the account from AI clients over MCP
Anomaly detection (deviation from pattern)
Incident auto-classification with precision shared
Predictive routing
Public model cards
Confidence scores in alerts
Customer data isolation (no cross-training)
GDPR Article 22 compliance
Tunable confidence thresholds
Some verticals get an outsized return on AI patrol. These are the three with strongest fit.
The AI agent in WhatsApp, Telegram, Slack and Teams is included from Free. When our AI models ship, they're part of the Business plan at no extra cost. If AI is core to the product, it belongs in the base price — not a surcharge.
For teams getting started
$0 forever
Standard guard tour platform
$29 /month
Full platform with AI modules on the roadmap
$79 /month
Pricing matches our hub at patroltech.online/en/pricing. Controllers (guards) are unlimited; billing is per checkpoint. Final amount calculated at checkout, no regional markups.
Three production models use machine learning: Anomaly Detection (Isolation Forest + temporal LSTM), Incident Classification (fine-tuned RoBERTa), Predictive Routing (constrained optimization with learned weights). Each has a public model card with training data, architecture, and evaluation. Two more in development. We don't call dashboards 'AI'.
Yes. The PatrolTech AI agent works in WhatsApp, Telegram, Slack and Microsoft Teams, plus other integrations. It answers questions about rounds, suspicious scans and patrol history, and it creates and edits checkpoints and patrollers or reconfigures routes when you ask. It is included from the Free plan, enabled by default and replies in your language. To start on WhatsApp or Telegram, just message PatrolTech at +34 624 085 772: if you do not have an account yet, the assistant creates it and configures it with you in the conversation. On Slack and Microsoft Teams, your organisation's admin enables it.
Yes. With MCP access you can operate your PatrolTech account from Claude, ChatGPT or Gemini.
No. Customer data is isolated per tenant. Anomaly Detection trains a per-site model on your data only. The base classification models were trained on a curated public + synthetic dataset (documented in the model card). We never use your data to improve products for other customers.
Every output ships with confidence score and explainable factors. Confidence thresholds are tunable. AI suggests — humans decide on any operational action. False-positive rate is exposed in the dashboard. We treat AI errors the same as production bugs: postmortems, public changelog, model retraining.
By design. No automated decisions with legal/significant impact on individuals. AI flags + suggests; humans approve. Audit logs separate AI suggestion from human action. We have a Data Protection Impact Assessment template available under NDA.
Yes, per site. Default 75% confidence for anomaly alerts. Adjust 50% (high recall, more false positives) to 95% (high precision, may miss subtle anomalies). The dashboard shows precision/recall tradeoff for your settings based on past 30 days.
Anomaly Detection needs 4 weeks of patrol data per site to baseline pattern. Incident Classification works from day 1 (pre-trained model). Predictive Routing needs 6 weeks of historical incidents for site-specific risk scoring. Until baselines are ready, the dashboard transparently shows 'Learning phase: X weeks remaining'.
Currently no. All production models are self-hosted (open weights or trained from scratch). LLM-based incident drafting (in-development for Q4 2026) will use Anthropic Claude via private API with no data retention contract. Disclosed in changelog when shipped.
We run a public AI Issues tracker. Anyone — customer or not — can submit observed errors, ethical concerns, or feature gaps. We respond within 5 business days. Critical bias/safety issues get priority triage within 24h.
Drop your work email and we'll send the technical walkthrough link plus a sandbox model card.
No credit card · GDPR-compliant · No cross-customer training · Unsubscribe in one click.