Book a demo
AGENTS LIVE ACROSS 42 TRADE LANES

The Operating Stack for Autonomous Supply Chains

AI agents that plan, book, reroute, and reconcile your freight — grounded in your contracts, your carrier history, and the systems you already run.

AGENTS RUNNING1,284
EVENTS / SEC18,402
SHIPMENTS MANAGED96,117
EXCEPTIONS AUTO-RESOLVED94.6%
[01] — CONTROL TOWER

One console for every lane, agent, and exception.

Every decision ships with its evidence, the options the agent rejected, and a one-click override.

[02] — AGENT RUNTIME

Agents run the loop. Continuously.

01SENSE

Ingest every signal

Vessel positions, port congestion, EDI, rate feeds, and weather — normalized into one event stream.

02PLAN

Reason over your context

Agents score every feasible plan against your contracts, SLAs, and carrier history.

03ACT

Execute in your tools

Book, amend, file, and notify — directly in your TMS, ERP, and carrier portals.

04VERIFY

Learn from outcomes

Each result is checked against intent and written back to memory, so the next decision is sharper.

[03] — SPEND RECOVERY

Find the money hiding in your freight spend.

Agents audit every invoice, accessorial, and free-time clock against your contracts — then file the dispute for you.

$18.42M freight spend, Q3 · $10K per pixel

Auditing ~/invoices/2026-Q3

RECOVERED$0K
  • Ocean freight$7.84M
  • Air freight$3.12M
  • Rail$1.46M
  • Customs & duties$2.08M
  • Road & drayage$1.61M
  • Demurrage & detention$2.31M
Looking for money you can recover
[04] — CONTEXT LAYER + MEMORY

Agents are only as good as their context.

Portex keeps one context layer over your whole operation — contracts, rate cards, SLAs, carrier performance, and live network state. Versioned, permissioned, and queryable by every agent.

  • 01 · EPISODIC MEMORYEvery decision, outcome, and override is stored as an episode you can replay and audit.
  • 02 · SEMANTIC MEMORYWhat agents learn about lanes, carriers, and seasonality — refreshed from every outcome.
  • 03 · SHARED STATEOne consistent world model across the fleet, so no agent acts on stale data.

Context layer

v2,481 · synced 0s ago · 6 sources × 16 lanes

LIVE
  • EPISODES
  • LEARNED
  • CONTRACTS
  • RATE CARDS
  • SLAS
  • LIVE STATE
    [05] — BUILT FOR ENGINEERS

    Define agents in code. Ship them like software.

    A typed SDK, policies in version control, and CI-gated deploys. Guardrails and spend limits are declarative — no prompt spaghetti.

    • [A]TYPED SDK, LOCAL DEV, AND CI-GATED DEPLOYS
    • [B]DECLARATIVE GUARDRAILS, BUDGETS, AND APPROVAL GATES
    • [C]FULL TRACE ON EVERY RUN, EXPORTED TO YOUR OBSERVABILITY STACK
    [06] — SIMULATION

    Replay two years of freight before you go live.

    Run every agent against 24 months of your own lane history. Policy violations fail the build in CI — not a shipment in production.

    • [A]SPEND LIMITS AND APPROVAL GATES CHECKED ON EVERY DECISION
    • [B]SLA AND CUSTOMER-PROMISE BREACHES FLAGGED BEFORE DEPLOY
    • [C]COST COMPARED SIDE BY SIDE WITH WHAT YOUR TEAM ACTUALLY DID

    Replaying lane history

    reroute-agent · SHA → RTM · portex sim --replay 24m

    0.0s

    0 decisions · 0 violations

    Decisions per second0/s
    Run 1 · 24 months · SHA → RTM
      [07] — INTEGRATIONS

      Native to the stack you already run.

      Two-way connectors with field-level mapping. Agents read and write where your team already works — nothing to rip out.

      SAP TMERP / TMS
      ORACLE OTMERP / TMS
      FLEXPORTFORWARDING
      PROJECT44VISIBILITY
      EDI X12 / EDIFACTMESSAGING
      SNOWFLAKEDATA CLOUD
      DATABRICKSLAKEHOUSE
      NETSUITEERP
      KAFKAEVENT STREAM
      SLACK / TEAMSAPPROVALS
      [08] — CONTROL & SECURITY

      Autonomy, on your terms.

      You decide what agents do alone, what needs a person, and what they never touch.

      Approval gates

      Set thresholds by spend, lane, or customer. Above them, agents propose and a person approves in Slack or Teams.

      Hard spend limits

      Budgets per agent, per lane, per day — enforced by the runtime, not requested in a prompt.

      Complete audit trail

      Every input, decision, and action is recorded and replayable — ready for finance, compliance, or a carrier dispute.

      Your data stays yours

      Tenant-isolated context, role-based permissions, and SSO. Your contracts and history never train shared models.

      “We stopped staffing a night shift for exceptions. The agents catch a rolled booking before our forwarder even emails us — and they’ve already priced the alternatives.”
      VP GLOBAL LOGISTICS · FORTUNE 200 ELECTRONICS
      “The context layer is the unlock. Our agents negotiate with carriers using our actual contracts and our actual history — not a generic model’s guess.”
      HEAD OF SUPPLY CHAIN TECH · GLOBAL RETAILER
      −38%EXPEDITE SPEND IN THE FIRST 90 DAYS
      11 minMEDIAN TIME FROM EXCEPTION TO FIX
      99.98%RUNTIME UPTIME, LAST 12 MONTHS
      $214MFREIGHT SPEND MANAGED BY AGENTS
      [09] — ROLLOUT

      Live on your first lane in four weeks.

      Agents earn autonomy lane by lane. Your team stays in control at every step.

      1. WEEK 1

        Connect

        Plug in your TMS, ERP, and carrier feeds. Read-only to start.

        READ-ONLY
      2. WEEK 2

        Replay

        Agents work your last 24 months. You see what they would have done — and what it would have cost.

        SIMULATION
      3. WEEK 3

        Shadow

        Agents propose on live freight. Your team approves every action.

        HUMAN APPROVES
      4. WEEK 4

        Autonomous

        Switch on autonomy lane by lane, inside the guardrails you set.

        AUTONOMOUS
      [10] — GET STARTED

      Put agents on your network.

      Connect one lane, replay its history, and watch agents work it — before they touch production.

      • READ-ONLY TO START
      • REPLAY BEFORE PRODUCTION
      • AUTONOMY LANE BY LANE
      [→] — BOOK A DEMO

      See agents work your lanes.

      Tell us a little about your network. We’ll come back with a time and a replay of one of your own lanes.

      WHAT SHOULD AGENTS HANDLE FIRST?
      Goes straight to the Portex team.