Board

Data reaches Board's models through Data Readers, JDBC and an SAP connector, native links to Amazon S3, Azure and Google storage, REST API call steps inside Procedures, and a Cloud Connector that also writes back to on premises sources through a secure tunnel. The agents sit on top of that data rather than calling those systems. They query Datasets through the Board engine, recommend matches, mappings and corrections, and can pass an answer table to a Smart Import object, where nothing loads until a planner reviews and saves it.

Sourcehelp.board.com/docs/interact-with-board-ai-agentsread 2026-10-12

An Agent Orchestrator takes each request with the context of the Screen it came from, decides which agent fits and which steps the task needs, and can sequence work across agents so that one coherent answer comes back. The FP&A, Controller, Supply Chain and Merchandiser Agents sit behind it, and an Agent Manager supervises the network. No calculation runs inside an agent. Each data request goes through a Dataset to the Board engine, which applies the model's logic and security, and the agent then explains the result and proposes next steps.

Sourcehelp.board.com/docs/create-and-configure-board-ai-agentsread 2026-10-12

Agents answer from the customer's own Board Data Models through Datasets switched on for AI, which fix the Entities, Cubes and selections an agent may query, and the Board engine runs every aggregation on that persisted multidimensional model. Data Attribute descriptions tell an agent what an ambiguous Entity or Cube means, and each agent can carry up to five uploaded text files of definitions, business rules, mappings and process guidance. An agent works on Screens built on a single Data Model, and Flex Grid layouts cannot yet feed an AI Dataset.

Sourcehelp.board.com/docs/create-and-configure-board-ai-agentsread 2026-10-12

Agents recommend and explain, and the decision stays with the user. Policy driven guardrails set which actions each agent may take, which workflows it may work in and when a person must approve, and an agent reaches only the data, models and actions its user may. An answer table sent to a Smart Import object loads into the plan only after a planner reviews and saves it, and the Controller Agent's proposed corrections and account mappings come to a person for review.

Sourceboard.com/airead 2026-10-12

Agents inherit Board's four layer security model, so roles, feature rights, folder security and data security on Entities and Cubes limit what an agent can reach exactly as they limit its user. Sign in runs on Board credentials with a custom password policy or through a SAML2 or OIDC identity provider, with SCIM provisioning and OAuth clients for API access.

Data is encrypted at rest with AES 256 and in transit with TLS 1.2 or higher, and customers can bring their own key. Board Cloud holds ISO/IEC 27001, 27017 and 27018 certificates, SOC 1 and SOC 2 Type II reports on request and a public SOC 3 report, and Board sits on the CSA STAR registry at Level 1.

Sourcehelp.board.com/docs/regulatory-compliance-certificationsread 2026-10-12

Every agent answer carries an Answer tab and, where relevant, a Chart tab, a Steps tab showing the steps and actions the agent took to build it and a Data Source tab showing the data behind it. Conversation logs and session histories stay in the customer's environment. Administrators track API Credit use against the monthly entitlement in the Subscription Hub, and its audit logs of user, connection and system events can be filtered, exported or pulled through an Audit Log API into a SIEM.

Sourcehelp.board.com/docs/interact-with-board-ai-agentsread 2026-10-12

Within one session an agent keeps the thread of earlier prompts and builds on them. That context resets when the page refreshes, when the user moves to a Screen with another agent or none, or at log out, so nothing carries into the next session. Inputs, outputs and conversation logs stay in the customer's Board environment and session histories are kept there, though an administrator cannot export the full set in bulk. A user can export single answers as text or Excel and charts as images.

Sourcehelp.board.com/docs/interact-with-board-ai-agentsread 2026-10-12

Board Cloud runs on Microsoft Azure in a dedicated tenant for each customer, segregated from other tenants, with one or more production instances and optional sandboxes.

The customer picks a supported Azure region, from Ireland, the Netherlands, Frankfurt, Paris, Milan, Zürich, Oslo and London to Virginia, Iowa, Illinois, Texas, California, Toronto, São Paulo, Pune, Hong Kong, Tokyo, Singapore, Canberra, New South Wales, Victoria, Dubai and Johannesburg, and Board's Cloud Team can validate one outside that set. Daily backups serve disaster recovery and restore points replicate across regions.

Azure's abuse monitoring can keep samples of agent prompts and answers for review by Microsoft staff in the European Economic Area, and the agents need the Board 15 cloud, although Board itself also installs on premises.

Sourcehelp.board.com/docs/cloud-architectureread 2026-10-12

Four ready made agents ship for distinct jobs, and a developer picks one by name and version when configuring it in a Data Model. The FP&A Agent covers variance and root cause analysis, statement alignment, external drivers and narrative commentary.

The Controller Agent brings a validation rule assistant, intercompany reconciliation that matches balances with learned patterns, data collection insights, group account configuration, cash flow mapping and local account mapping. The Supply Chain and Merchandiser Agents, from release 15.1, analyze, explain and simulate options before a change or a buy. The Architect Agent is in preview.

Sourcehelp.board.com/docs/about-board-ai-agentsread 2026-10-12

A planner starts every agent session from the AI icon on a Screen the agent is linked to, by typing a question or choosing one of the suggested prompts a developer prepares, up to six groups of six. The agent takes the Screen's selections and filters as its starting context and answers inside Board. Nothing starts an agent without a person. Scheduled Procedures and email broadcasts run on their own, but no Procedure step calls an agent, and the agents do not reach people in email or a team chat.

Sourcehelp.board.com/docs/interact-with-board-ai-agentsread 2026-10-12

Board Agents run on OpenAI language models served through Microsoft Azure, and a customer cannot move them to another provider or its own model. What a customer does control is the agent version, chosen from recommended, supported or deprecated releases and upgraded only when the customer decides. Bring your own model in Board means predictive forecasting models, brought in through Azure ML Studio, Databricks, R, Snowflake or REST APIs, not the model behind the agents.

Sourceboard.com/ai-policyread 2026-10-12

Outside systems call Board through its Public APIs, REST endpoints that return API Query results as JSON, browse Capsules, Presentations, Slides and Screen objects, search the platform, list a Data Model's Entities, members and Cubes, and start a Data Model Procedure with a POST. Access runs through OAuth clients an administrator creates in the Subscription Hub, with scopes and rotating secrets, beside SCIM APIs for users and groups and an API for audit logs. The agents themselves have no API, and outside assistants cannot reach Board through an MCP server.

Sourcehelp.board.com/docs/board-public-apisread 2026-10-12

Each answer shows its steps and source data so a user can check it, and Board asks users to validate every answer. Agent versions stay pinned until a developer upgrades, so a new release changes nothing on its own. No test set or measured evaluation run checks how an agent answers before a team relies on it. The Architect Agent, in preview, assesses a planning model's dependencies before a change, which checks the model rather than the agents.

Sourceboard.com/ai/architect-agentread 2026-10-12

Board Agents query Datasets that the Board engine computes, and an answer table can go to a Smart Import object for a planner to review and save. No agent opens or drives another application's screen, a browser or a desktop session for a user.

Sourceboard.com/airead 2026-10-12