Beta — the agent-first release

Built by safety engineers, for safety engineers

Agent-First
System Safety.

Your AI agent reads the hazard log, drafts the summaries and change requests, and SafeForge keeps every change human-reviewed and defensible — from import to assurance, with live bow-ties, risk workshops, and DOORS & SharePoint connectors.

Beta: 50% off until beta ends with code FORGEDSAFE at checkout. Want to be a beta tester? Email support@safeforge.ai →

Hazard: Signal Passed at Danger (SPAD)

Key barrier states drawn from the findings of the Cullen Inquiry (2001); some barrier detail added for demonstration. Hover any cause, barrier, or consequence for detail. Greyed, crossed barriers were absent or rejected before the event.

About this hazard

A SPAD occurs when a train passes a stop signal without authority — one of the classic precursors of rail collision. The causes span engineering, environment, and human performance; most SPADs end as near misses, which is exactly why weak signals accumulate quietly in the data until the barriers align against you. SPADs drove three major UK accidents in a decade: Purley (1989), Southall (1997), and Ladbroke Grove (1999).

The real-world incident behind this diagram

This bow-tie summarises a real accident: 31 lives were lost at Ladbroke Grove. The inquiry found the gaps were visible beforehand — 8 SPADs at signal SN109 in six years with no effective corrective action; signal-sighting committees never convened around Paddington since 1994 despite two audits; and ATP rejected on cost-benefit grounds after being recommended following Clapham Junction (1988). Diagram simplified from the public findings, for education.

This is the purpose of SafeForge — we help you forge Safe Systems.

Agent Import Bow-Ties Workshops Assist Governance Connectors
SafeForge Agent Beta

Ask. It reads the log. You keep the call.

Work in SafeForge, or in the AI assistant your team already uses — via MCP. SafeForge Agent reads your hazard log through governed tools, drafts summaries and change requests, and the app follows the conversation on screen. Every draft lands unconfirmed; every change waits for a reviewer.

  • Follow mode — the app goes where the conversation goes
  • Governed checks — findings against good international practice, never a verdict
  • Change requests raised by the agent, accepted by a person
The rules the agent lives by — read the SafeForge AI Dogma →
SafeForge Agent
connected to Mt Verdant Tramway · illustrative exchange, synthetic project

Did we miss a cause on the Train derailment key risk?

Two candidates from published derailment investigations are in your cause library but not on this key risk: brake fade on the long descent (THR-014) and wheel-flange climb on the tight curve (THR-021). Both appear in investigation reports for tramways with a comparable gradient and curve radius. Want me to run a research pass on either?

↳ read key-risk bow-tie↳ searched published investigations

Yes — brake fade. And drop the manual points inspection sheet; the points monitoring system replaced it last year.

Staged on the key risk, AI · unconfirmed until you confirm: THR-014 Brake fade on long descent with Speed supervision on the descent (CTRL-P-038) as its preventive control, and CTRL-P-027 Manual points inspection sheet removed from this bow-tie. Key-risk edits need no change request — you confirm them.

One thing to know: removing it here does not retire it. CTRL-P-027 is still an active control in your hazard log, on two other hazards. Retiring it there is a change request, and a retirement needs a justification and evidence — what replaced it (CTRL-P-003 Points monitoring & lock?), from what date, and the inspection records covering the overlap. Want me to draft CR-013 from the audit trail for you to complete?

↳ stage ×2 · unconfirmed↳ read control CTRL-P-027 · linked hazards

Advisory — a comparison to good international practice, not a legal determination. An AI assistant can make mistakes; a person confirms or rejects every proposal in SafeForge.

Smart Import Live

Bring your whole log across in minutes.

AI maps your columns on first import — Excel, CSV, or DOORS exports — then remembers the format. Re-imports are instant and free, so your existing hazard log lands intact, not retyped.

  • AI column mapping that learns your format
  • Any industry format — Excel, DOORS, CSV
  • Free, instant re-imports once mapped
Interactive Bow-Ties Live

Every risk, mapped end to end.

Causes on the left, the top event in the centre, mitigations on the right — live, clickable, and stakeholder-ready. AI drafts grouped summaries of your key risks when you need to brief upwards.

  • Real-time editing — every box clickable
  • Preventative & mitigative barriers on one canvas
  • AI-drafted risk summaries for stakeholders
Explore the live SPAD case-study bow-tie above ↑
Risk Workshops Live

A copilot in the room — the room keeps the pen.

Pre-populate the sheet from your frame and your own control library, ask the agent whether you missed a cause, let it research published incidents and propose controls with their sources — then remove, alter and own each item as a room. Lock what you agree, share the live bow-tie on the wall, and close out into the log through review.

  • Every AI item stays flagged until the room confirms it
  • Research passes cite sources for the room to verify, never to obey
  • Live-shareable bow-ties; close-out raises a change request
Assist Live

A health check across your whole log.

Assist reviews the log the way a seasoned reviewer would — gaps in control coverage, orphaned causes, weak linkage — and compares your controls to good international engineering practice. Every finding is a suggestion; your team decides what enters the log.

  • Gap and coverage findings across the log
  • Comparison against international practice — with the caveats stated
  • Human review on every AI suggestion
The rules our AI lives by — read the SafeForge AI Dogma →
Governance Live

Defensible by Design — without the friction.

Two-person review with per-entity approval, baselines, and a full audit trail. Define V&V rules for control-requirement linking and SafeForge flags violations automatically during review.

  • Two-person change review, per-entity approval
  • Baselines and complete CR history
  • Automatic V&V rule-violation flags
Connectors Live

Update your log live from DOORS or SharePoint.

Point SafeForge at a DOORS module or a file on SharePoint and your hazard log stays current — changes flow in through the same review gate as everything else, so the log is live and controlled.

  • Live sync from DOORS modules
  • Watch a SharePoint file and pull updates
  • Scheduled exports back to SharePoint — your log auto-backs up
  • Every change still passes two-person review

Import it, workshop it, or ask the agent. AI drafts. Your team reviews. Nothing enters the log without a human.

One subscription. The whole toolkit.

Everything above, plus the working parts below — no per-module pricing. Compare plans

  • Agent access via MCPThe same governed tools from the AI assistant your team already uses, over MCP — read, check, propose; a person accepts.
  • Bonus read-only seatsOne free read-only seat with every paid seat — executives, client reviewers and auditors follow the log, the bow-ties and every change request without being able to alter them.
  • TemplatesHazard-log and risk-worksheet Excel templates that re-import without AI mapping.
  • ExportsProfessional export templates with locked structure, ready for stakeholders.
  • Audit trailBaselines, CR history, and a complete record of who changed what, when.
  • V&V rulesYour verification & validation rules, enforced automatically during review.
  • DashboardStatus distributions and key risks at a glance across the project.

Built for Safety-Critical Industries

🚂

Rail

DOORS, EN 50126, hazard logs

🛡️

Defence

DEF STAN 00-56, MIL-STD-882E

✈️

Aviation

ARP 4761, FHA, PSSA/SSA

💧

Oil & Gas

HAZOP, ISO 31000, bow-tie

Download Our Templates

Professional Excel templates built on the SafeForge risk model. Use standalone, or import straight into the app — no AI mapping required.

📃

Hazard Log Template

Multi-sheet workbook with hazards, controls, causes, assumptions, and requirements. Pre-formatted with dropdowns, conditional formatting, and the full SafeForge risk model.

  • 5 register sheets, Excel Tables with auto-filter
  • Data validation on status, category, severity
  • RAG conditional formatting on risk columns
  • Re-imports directly to SafeForge
Download (.xlsx)
📋

Risk Worksheet Template

Branded worksheet for offline hazard identification workshops. Colour-coded zones for hazard, cause, initial risk, controls, and residual risk. 30 blank rows pre-formatted.

  • Single-page register, print-friendly
  • Dropdown validation on type, severity, likelihood
  • Editable workshop metadata (facilitator, participants)
  • Additive re-import — creates new entities
Download (.xlsx)

Simple Per-User Pricing

All features included. Every paid seat comes with a bonus read-only seat — for the people who need to see the record, not change it. AI credits pooled across your organisation. Prices in USD.

Beta pricing — 50% off until beta ends with code FORGEDSAFE at checkout. Want to be a beta tester? Email support@safeforge.ai

Stop wrangling spreadsheets.
Start managing risk.

One platform from import to assurance. Your data stays yours; your reviewers stay in charge.

All features included · bonus read-only seat with every paid seat · 14-day free trial · cancel anytime