Frequently asked questions

NeoXam Agents is a unified AI agent layer for investment management operations, embedded across the NeoXam product suite (ARO, DataHub, GP, Impress, PMS). Agents are bounded, declared capabilities — investigation, drafting, explanation — invoked from the products your teams already use, governed by a Git-backed catalog, evaluation baselines, tool policies, and full tracing.

Access is provisioned per organization while the platform is pre-GA: NeoXam clients in the Early Adopter Program and selected pilot teams. Your workspace administrator provisions your work email; there is no public self-service sign-up. Sign-in is a passwordless magic link today, with enterprise OIDC SSO rolling out in the V1 Beta.

No. NeoXam never fine-tunes models on your data, and no business data feeds back into any model. Models are used for inference only. The platform runs in the EU (Frankfurt) by default, your data stays inside your tenant boundary, and credentials never appear in prompts, logs, or traces.

General availability of the unified platform is planned for Q3 2026. The V0 platform is delivered — task-mode agents, including the ARO reconciliation pilot — and the V1 Beta (conversational assistants, human-in-the-loop approvals, OIDC SSO) is in build now.

The program is open now as a limited cohort: a one-year platform trial with three workshop streams — Business ROI, Compliance, and Operational fit. Request access through the contact page and our team will follow up.

Today the platform runs Anthropic Claude through an AI gateway. The architecture is model-agnostic by design — the model is a configuration field in each agent's descriptor — so switching or mixing providers is a configuration-and-re-evaluate operation, not a re-architecture. Models are never fine-tuned on your data.

The platform is SaaS in the EU (Frankfurt) today. Hybrid, on-premises, and sovereign deployments on Kubernetes are committed for GA (Q3 2026), with a self-hostable observability stack so traces stay inside your boundary.

Commercial terms are defined with Early Adopter Program participants and depend on deployment and usage. Cost is bounded by design — per-run budgets, per-tenant quotas, and cost dashboards. Contact us for a conversation about your scope.

The Console is a web application supporting current versions of the major evergreen browsers — Chrome, Edge, Firefox, and Safari. No plugin or local installation is required.

Enterprise OIDC SSO — compatible with KeyCloak, Azure Entra ID, and Okta-class providers — is rolling out with the V1 Beta. Today, sign-in uses passwordless magic links with full token verification.

Every run is reconstructable step by step: the exact catalog version executed (Git SHA), the model, ordered tool calls with inputs and outputs, tokens, cost, and duration. The audit log is append-only, designed for 7-year retention, and exportable as CSV for compliance. Human approvals and denials (Beta) are recorded with who, when, and why.

Only within what their declared policies allow. Every tool call passes an allow / ask / deny policy; irreversible actions default to ask or deny, and ask pauses the run for an explicit, recorded human decision (rolling out with the V1 Beta). Agents never orchestrate business workflows — the application decides what runs and when.

The NeoXam Agents Console ships in English first. This website is available in English, French, German, Spanish, and Italian. Localization of product surfaces is gathered as feedback through the Early Adopter Program.

No. Agents do the digging — investigation, drafting, explanation — within bounded tasks, and return structured findings with evidence. Your analysts review and decide, deterministic systems keep the books, and sensitive actions require human approval. The platform is built for oversight, not replacement.