Get Pigment AI working on your model.

Pigment AI is a set of agents that work inside your Pigment workspace. Four are available today, and Pigment also offers an MCP Server for outside AI tools:

• Modeler Agent: builds and edits Blocks, formulas, Views and Boards from a plain-language request, in Ask, Plan & Build or Build mode
• Analyst Agent: answers questions about your data, drills into drivers, writes reports and runs repeatable Missions on a schedule
• Custom Agents: agents your team builds with its own instructions, knowledge and skills
• Documentation Agent: answers questions about how Pigment works, drawing on Pigment’s own documentation, and explains formulas
• Pigment MCP Server: lets tools such as ChatGPT, Claude and Microsoft Copilot query Pigment, within each user’s permissions

Because an agent can only reason about what the model tells it. Pigment’s own guidance says Block names, descriptions and formula annotations directly affect how well its AI performs. Undocumented, ambiguous Blocks lead to less precise suggestions, and agents inherit the access rights of the person using them.

We check the things Pigment’s agents depend on, then fix them or show your team how:

• Naming: technical and display names a person and an agent can both read, with no vague labels
• Descriptions: a one-sentence purpose on every Block, saying what it holds and whether it is input, calculation or output
• Annotations: formulas explained in plain language, so the Modeler Agent follows your logic
• Access: role permissions for using and configuring agents, and AI data access set Metric by Metric
• Data shape: time aggregation on Metrics such as headcount or FX rates, so the Analyst Agent doesn’t add up things that shouldn’t be added up
• Governance: where agents may build (development, not production), who reviews their plans, and how AI credits are watched

Start with the Modeler Agent in a development application, ask it to plan before it builds, and have a person review every plan. Grant agent permissions by role, keep web search off or limited to approved domains, and turn AI data access off for Metrics that shouldn’t be queried. Then widen use as the team gains confidence.

Yes. We design Custom Agents around a specific job, such as answering budget-owner questions or explaining variances. That means clear instructions, a small set of good knowledge sources (process notes, key Boards, a glossary), the right skills, and testing with the questions your team really asks.

Frames are custom interfaces built on live Pigment data, in open beta since September 28, 2026. You describe what you want and the Modeler Agent writes it, or you build from Claude, ChatGPT or Cursor through Pigment’s MCP Server. A Frame reads and writes the live model and respects each person’s access rights, so a regional manager and the CFO see different numbers in the same Frame. We design and build Frames where Boards fall short, and set the rules for who reviews and publishes them:

• Guided input forms and approval flows for budget owners who rarely open Pigment
• Executive views and narrative reports laid out the way leadership reads them
• What-if tools, such as reallocating sales quotas or testing a hiring plan
• Knowing when not to: Boards remain better for dense grids, exports and large data volumes

Yes, through the Pigment MCP Server. We help your security administrator enable it, connect the AI tools your company already approves, and decide which Metrics those tools may read. External tools can read only number and true/false Metrics, and only where AI data access is switched on.