What Atlas is
Atlas is the in-product AI working inside TradeOS on the operator’s actual deal context. Not a chatbot bolt-on. Atlas reads from the same data model the rest of the platform writes to, with full context on every entity, every transaction, every relationship. When Atlas drafts a contract, it knows which manufacturer it’s for, what the on-time delivery rate has been, what the dispute history looks like, which jurisdictions apply, and what the operator’s standard terms typically are. No general-purpose chatbot can do this. No bolt-on integration of standalone tools can fake it.
The four areas
Legal AI
Contract drafting, redline analysis, risk flagging, jurisdictional sensitivity. Sale agreements, supply agreements, NDAs, master services agreements. Works from the operator’s actual deal context (counterparty, performance history, jurisdiction, standard terms) rather than a generic template.
Accounting AI
Multi-currency reconciliation, partial payment matching, split-allocation handling, LC structure parsing, automatic chart-of-accounts mapping. Replaces the spreadsheet-and-coffee reconciliation cycle that occupies finance teams every month-end.
Predictive AI
Late delivery flags before they materialise. Capacity warnings before the manufacturer raises them. Tariff alerts in destination markets. OKR tracking against actual pipeline reality. Foresight grounded in operational data, not external forecasts.
Automation
Repeatable workflows fired on conditions. Document chasing, payment reminders, customs declaration generation from order data, QC inspection scheduling, exception-routing to the right human. Executes in the background while the team focuses on judgment work.
Multi-provider AI floor
The target accuracy curve is 82% in month one, 94% in month two, 98% by month six as the system learns the operator’s specific patterns, vocabulary, and exception rules. Atlas is built on a multi-provider AI architecture spanning Claude, OpenAI, and Gemini, with a local Gemma deployment as a resilience floor. The platform keeps working if any single provider is unavailable.
The multi-provider design is not for cost optimisation. It is for resilience and sovereignty. Operators running cross-border trade flows cannot afford a stalled platform because one AI provider had an outage or a contract change. The local Gemma deployment ensures the platform continues to function even if every external provider is unreachable at once.
Three data sovereignty tiers
Standard
Zero data retention contracts with all providers (Claude, OpenAI, Gemini). Operator data is processed by the provider for the duration of the request and not retained. The default tier for most operators, ships with every paid TradeOS subscription.
Sovereign
Confidential computing in the customer’s own cloud VPC. No external provider ever sees raw data; processing happens inside the customer’s cloud boundary with attestation-verified compute. The tier for operators with strict data residency or regulatory obligations.
Air-Gapped
Fully self-hosted on customer infrastructure with the local model only (no external provider used at all). The tier for operators in sectors where outbound data flow to any AI provider is prohibited (defence, sovereign customs, regulated public-sector trade).
Where it stands today
Atlas is in active development. Professional Services AI (Legal AI and Accounting AI) ships at TradeOS public launch (August 1, 2026), at the end of Stage 1. Automation and Bots ships in MVP as section fourteen. Predictive AI and Strategic Management ship through Q4 2026 and Q1 2027 in parallel with the L2 mainnet and Global Trade Marketplace launch. The multi-provider AI architecture is already operational on the platform; the data sovereignty tiers ship in stages, with Standard at launch, Sovereign in Q4 2026, and Air-Gapped in Q1 2027.
Pricing at edma.trade. Full roadmap at /roadmap/.
