The Agent Mesh

Meet the agents.

Agents at Proxima360 are not chatbots with a retail vocabulary. Each one has a beat it watches, data it learns from, arithmetic it shows, and actions it can take inside your guardrails.

๐Ÿ† The Roster ยท The Ceremony

Ten agents. One platform.
The roll call.

Roxi takes the stage and calls the roster โ€” every agent, every city, every beat, one signature line each. This is the team, introduced the way a team should be.

Watch Agents ยท learning from daily snapshots
๐Ÿšข
Movement Agent
deliveries vs promises ยท per supplier

Finds chronic slippage, quantifies it, tells you how much to pad.

movement.insight โ†’ "Ningbo Cordage delivered late in 6 of 10 POs, avg slip 5.2 days โ€” pad ETAs."n=10 observations ยท confidence MED ยท arithmetic printed
๐Ÿ”ฌ
Quality Agent
inspection history ยท defect drift

Flags defect drift before it becomes a failed shipment; recommends sampling posture per supplier.

quality.insight โ†’ "Defect rate trending 2.8% โ†’ 4.4% over 3 inspections โ€” tighten sampling."n=9 ยท confidence MED
๐Ÿ›ƒ
Customs Agent
entry variance ยท statutory clocks

Surfaces overpayment patterns and urgent deadlines before recoverable money expires.

customs.insight โ†’ "$19.4K net overpaid across 6 entries โ€” 1 protest clock inside 30 days."n=6 entries reconciled ยท confidence HIGH
๐Ÿ’ฐ
Cost Agent
estimate vs invoice calibration

When reality drifts from the model, it says by how much and where.

cost.insight โ†’ "Freight calibration drifted +9% over 28 days โ€” model updated from invoices."n=28 daily snapshots ยท confidence HIGH
๐Ÿ“ˆ
Plan Agent
open-to-buy pace vs plan

Flags departments burning too fast or too slow against the plan.

plan.insight โ†’ "Commitment pace 72% with 3 months left in window โ€” ahead of plan."n=28 ยท confidence HIGH

Every insight: printed arithmetic, sample size, confidence tier. Under 5 observations, an agent says "still learning" โ€” and that honesty is a feature, not a disclaimer.

Research Agents ยท web-grounded intelligence

Account intelligence in three phases.

01

Price the deal

Public financials, scale and tech spending โ†’ a priced recommendation with win probability. Seconds, not weeks.

SECONDS ยท AUTO
02

Deep dossier

Market position, leadership changes, competitors' moves, ownership & PE activity, tech estate, incumbent vendors & SI partners, hiring signals โ€” three parallel research passes.

~2 MIN ยท AUTO
03

Very deep research

Job postings triangulated into an investment map โ€” every inference chained to its evidence. Trigger events, buying window, buying committee, trade posture, value hooks in their own reported dollars.

ON DEMAND

Facts are sourced. Inferences show their chain. What wasn't found is listed โ€” not invented.

Execution Surfaces ยท where agents act
๐Ÿงพ

Draft & hold

POs drafted from priced alternatives straight into the pipeline. Spend above your threshold holds itself for audited sign-off โ€” desktop or phone.

๐Ÿ“ก

Track & assemble

Carrier webhooks move shipments forward-only through the pipeline. CBP-ready dossiers assemble with an honest gaps list.

๐Ÿ””

Alert & recover

One bell unifies every urgency with deep links โ€” capped and deduplicated so it stays meaningful. Recovery claims run against statutory clocks.