Tech Talks: SensibleAI Agents Replay now available!

Curious about SensibleAI Agents but not sure where to start? Join OneStream Senior AI Consultant Suren Lad, Tom Linton, and Kayne Schwarz as they demystify OneStream's AI capabilities, walk through a real implementation from setup to deployment, and show how Finance Analyst and Search can deliver value with no custom coding required.

SensibleAI Agents

• What are SensibleAI Agents? SensibleAI Agents is a paid, separately licensed capability within the OneStream platform that gives users a natural language interface to their financial data and documents. Three agents are available: Finance Analyst (query cube data via natural language), Search (query uploaded documents and connected document sources), and Deep Analysis (covered in a separate session). All are accessed through a new slide-out panel in the top-right corner of the OneStream UI.
• Purpose-built agentic layer: SensibleAI does not connect an LLM directly to your OneStream data. Instead, OneStream maintains a dedicated agentic layer that sits between the LLM and your cube data. This layer understands OneStream's dimensional structure, enforces your native security, and ensures the AI stays focused on the right data — regardless of whether you access it through the native interface or via MCP from an external tool like Claude or Copilot.
• Security is preserved: The agents fully respect the security groupings already configured in OneStream. Users cannot access data through the agent that they cannot access in a Quick View or Cube View. A second layer of control allows admins to further restrict which members and dimensions the agent can even "see" — useful for keeping work-in-progress scenarios or sensitive data out of agent scope entirely. Important note: once data is surfaced to an external LLM via MCP, it is governed by that provider's own data policies.
• Finance Analyst in action: Ask a question in natural language (e.g., "Show me North America equipment product line trends over the last six months") and the agent builds dynamic Cube Views on the fly, returns a natural language summary with source-linked numbers, and cites its sources. Clicking any number opens the underlying Quick View or Cube View for verification. Data is auditable and traceable back to the cube.
• Finance Analyst Plans: Recurring analysis schedules that automatically run a defined set of questions and produce a formatted report — useful for month-end close packs, flash reports, or op-ex reviews. Plans are time-based (daily, weekly, monthly) and can also be triggered manually. Workflow-based triggers are on the roadmap.
• Personas: The mechanism for tailoring agent behavior to different user groups or business divisions. Each persona can have its own configured cubes, dimension defaults, member filters, Cube View access, and business context. A top-level "company" persona holds shared settings; individual personas inherit company context and add their own. Users can be assigned a default persona by an admin.
• Cube sync and business context: Before Finance Analyst can answer questions, cubes must be "ingested" — a sync that imports your metadata into the agentic layer. Resync is needed only when metadata changes, not after data loads. Finance analyst capabilities are temporarily offline during a sync. Business context is a free-text natural language field (up to 20,000 characters) at the company and persona level that gives the agent institutional knowledge: how to interpret members, which cube to use for which question, division-specific terminology, and commentary write-back rules.
• Document clusters for Search: Documents are organized into clusters (e.g., "Core OneStream Books," "Accounting Policies") that can be uploaded directly or synced from SharePoint or Confluence. Cluster access is secured separately, so HR documents and finance documents can be restricted to the right users. All cluster changes require an approval step — even when an admin creates them — providing audit-friendly change control.
• Credits and AI Control Tower: Token usage is managed through a credit system visible in the AI Control Tower. Admins can view credit history, see usage by user and by agent type (native vs. MCP), set per-user credit limits to prevent runaway usage, and monitor overall consumption. Credits are consumed on a per-question basis, with complexity affecting the cost.
• Getting started: SensibleAI Agents requires a paid license, a support ticket to have OneStream deploy the agentic layer to your environment, and installation of the marketplace workspace. A dedicated server user (Nikki Server) is provisioned as part of setup and counts toward your user license. First-time partner implementations require at least one OneStream advisor on the project.
• Key Takeaway: SensibleAI Agents democratizes access to your OneStream financial data by enabling natural language queries — but it requires deliberate configuration to be effective. Time spent training personas, defining business context, managing dimension members, and curating Cube View access translates directly into the quality and accuracy of answers users receive. Think of the setup as onboarding a highly capable new analyst who needs your institutional knowledge documented before they can be trusted to work independently.

To receive OnePoints for watching this Tech Talk video, please visit the link below:

Navigator link: Tech Talks: SensibleAI Agents