Platform
The operating model
is the product.
AI models and chat interfaces will commoditize. What endures is a typed freight ontology, governed actions, provenance on every fact, and privacy-preserving network intelligence — one core, progressively revealed.
The freight graph
Typed objects. Stable identities. No competing copies.
The canonical model represents freight once. A draft movement can hold zero legs; dispatch requires one or more; the same identity later composes handoffs, jurisdictions, currencies, permits and settlement events. No module may create its own version of a shipment because its screen looks different.
Events are append-only. Corrections supersede or qualify prior assertions — they never erase evidence.
Sixteen core types, one graph — extended by versioned capability packs, never forked.
Living workspace
Progressive configuration, four gates, zero surprises.
A capability appears only when it is entitled by your plan, permitted by policy, configured with the facts it needs, and relevant to your actual operation. Learning creates proposals — an authorized person approves, edits, defers or rejects.
Now
Execute
Observe
Coordinate
Since your last shift
Refreshed 06:12 CAT
2 loads delivered; 1 still needs POD review.
FL-204 may miss the 16:00 window — last confirmed position 13:42.
A carrier reported a border delay on FL-198; no independent update yet.
Workspace proposal — border handoff tracking
Evidence: 3 cross-border movements in 30 days. Nothing changes until you approve it.
Qualified truth
Workflow state and truth state never blur.
“In transit” is a workflow state. “Driver reported 14:08” is evidence qualification. “ETA 16:25 at 72% confidence” is a prediction. Fraytline renders all three distinctly, everywhere — boards, maps, portals, answers — and keeps full lineage one action away.
GPS ping, geofence event, scanned barcode — machine-witnessed.
A person or partner said so — identity verified, source retained.
Extracted or derived — method and region referenced.
A model's forecast — confidence attached, never dressed as fact.
An authorized actor committed it — attributable, auditable.
Intelligence runtime
AI-native is not agent-heavy.
One governed runtime routes every task to the cheapest sufficient method — rules, exact queries, statistics, templates, caches — and invokes a bounded specialist skill only when justified. No standing agent per shipment, truck, user or document. Operational state lives in the freight model, not in an agent's memory.
1 · Deterministic
Rules, validation, exact lookups. Free, instant, correct — the default path.
2 · Retrieval
Permission-scoped search over authorized objects and evidence only.
3 · Bounded skill
A specialist model task with budget, timeout, and typed output.
4 · Receipt
Initiator, sources, scope, method, cost, confidence, resulting action — always.
Provider outage or budget stop? Core freight execution continues — deterministic paths never depend on generative AI availability.
Security model
Three surfaces. Hard boundaries.
Operations, Company Settings and platform control are separate applications with separate routes, queries and policies. UI hiding is not authorization — every query, export, subscription and AI retrieval enforces the same boundary server-side.
| Concern | Operations | Company Settings | Fraytline Control |
|---|---|---|---|
| Freight records | Authorized daily work | No implicit access | No standing access |
| Users & roles | View collaborators | Manage within scope | Identity health only |
| Tenant AI | Answers & feedback | Allowance & features | Runtime policy, redacted telemetry |
| Support access | Visible indicator | Approve / revoke | Reason-coded, time-boxed, audited |