The AI platform built for
how your firm actually works.

Multiple people in a single AI conversation in real time. One inference call serves all participants — five people cost the same as one. Five roles: Owner, Collaborator, Observer, Supervisor-Visible, Supervisor-Silent.

Standard enterprise AI subscriptions are single-player. One shared thread means one governed, auditable record — not five parallel conversations nobody can reconcile.

Fork any conversation at any point. The branch inherits full context from the fork point. Classification can escalate from the parent but never de-escalate.

Explore an aggressive position on a branch while the main line stays conservative. Every branch runs through the same governance pipeline.

Switch between Claude, GPT, and other credentialed models mid-conversation. Full context is translated server-side. Model switching respects classification policy — restricted tiers stay within admin-approved models.

Single-vendor AI subscriptions lock you to one model family. NuBench routes each workload to the best-fit model — without losing governance, audit trail, or context.

Three optimization modes: Performance, Efficiency, and Balanced. Efficiency mode reduces token cost by summarizing context — users pick the mode, admins set the ceiling by role or hierarchy level.

Teams run heavy workloads efficiently. Admins control spend by role without micromanaging every conversation.

Tag every AI interaction to a client engagement before it runs. Three billing modes per cost code: Absorb, Allocate, and Pass-Through. Exports to your PSA via CSV or API.

Standard enterprise AI subscriptions have no per-matter attribution. Client work gets billed to the right cost code automatically — no manual reconciliation at month-end.

Persistent teams of AI agents deployed into your organizational hierarchy. Every governance rule that applies to human users applies identically to fleets — classification, budget, and audit included.

Agents are not a shadow IT problem. They live inside the same policy boundary as your team — the same DLP, the same spend limits, the same audit trail.

Six configurable levels. Knowledge flows down — every KB, template, and config defined at a parent level is available to every level below. Costs roll up automatically. Budget caps cascade.

The org structure your firm already uses maps directly to AI access, budget, and context inheritance. No second hierarchy to maintain.

Do software development work in the same governed, billable, classified way as every NuBench conversation. VS Code extension connects to the same 16-step gateway.

Every accepted completion is attributed to a cost code and recorded in the audit trail. Developer AI spend is visible and controlled like every other AI spend.

Our cost attribution engine enforces token budgets in real time, before the model is called. Per org, per team, per user, per matter — overage is impossible, not just visible.

No surprise invoices. Finance sets a cap per matter and it holds. Teams spend confidently because they know when they're close — not after the invoice arrives.

Each message is classified and scanned before it leaves your boundary. Flag, redact, or block — your policy, applied in real time, before any model sees the content.

Client confidentiality is enforced before the model call, not logged after. Staff use AI for sensitive work without risking accidental disclosure.

Every prompt, response, policy decision, and admin action is written to an immutable, hash-chained log with export built in.

Evidence of what AI was used for, by whom, on which matter. Answers the regulator's question and the partner's question from the same record.

SAML/OIDC single sign-on with seven platform roles. Nobody shares an API key, and access ends the moment someone leaves.

Onboarding is a directory sync. Offboarding is instant. No orphaned sessions, no shared credentials.

OpenAI, Anthropic, Google, or Azure — your keys, your data-processing terms, your negotiated rates. One governed interface for all of them.

AI costs bill directly to your provider account — no per-seat markup, no opaque bundling. Rotate a key and access stops immediately. Your existing enterprise pricing applies.

Your data is isolated and encrypted at rest. Enterprise plans add dedicated single-tenant infrastructure and customer-managed encryption keys.

Standard enterprise AI subscriptions share infrastructure across all customers. Your data never shares compute, storage, or network with another organization.

Multi-model access

Claude and GPT via your org BYO keys, plus Amazon Bedrock defaults (Nova, Llama via Bedrock). Gemini via Admin BYO when credentialed — Vertex for Confidential/Restricted content; Google AI Studio limited to General (PUBLIC/INTERNAL). Mistral via Admin BYO. Classification and credential shape determine eligibility. Mid-thread model switching is confirmed; a Jul 2026 Gemini “live” demo was a COM routing banner only and is retracted (not Google dispatch).

Projects and workspaces

Organize conversations, files, and team members into projects scoped to a client engagement or workstream. Context persists across every conversation in the project. Classification is enforced at the project level.

Persistent memory

User-level and org-level memory. Classification-scoped — restricted tiers inherit the same policy boundaries as chat. Visible, editable, erasable, and exportable by the user.

Knowledge base and RAG

Upload documents or connect Google Drive and Confluence. Classification-aware and hierarchy-aware retrieval. Knowledge defined at a parent level cascades to all levels below.

Skills library

Pre-built and custom AI workflows invoked by name. Every skill execution is logged. Custom skill builder follows in Phase 2.

Conversation templates

Pre-loaded context, model selection, and tool configuration by task type. Every published version is immutable — downstream integrations never break from template drift.

MCP server support

Admins configure Model Context Protocol server endpoints. Classification-scoped. Internal tools are available to skills and agent workflows.

Shared conversations

Async sharing within hierarchy scope. Named users receive read or comment access. Classification enforcement applies identically to all shared content.