Customize an Agent

Customizing an AI agent allows you to move beyond the out-of-the-box agents and create a specialized digital worker that understands more about your specific project logic, standards, or reporting requirements. This tutorial is for the following Datagrid AI audiences:

• Datagrid customers wanting to optimize AI results after getting started in the Datagrid standalone product.

• Procore customer wanting to optimize AI results after getting started in Procore.

• Prerequisites:

  • Create an agent in Procore or in Datagrid's standalone product.

  • An Enterprise-tier Datagrid subscription is required to create and maintain custom global agent templates via the Datagrid standalone product.

  • Procore customers: To customize an agent in Procore or Datagrid Standalone that then also shows up in the Datagrid AI side panel in Procore, your account must be part of an active Datagrid AI rollout cohort with Datagrid enabled in the Company Admin tool.

• Visibility for Procore customers: If your account has Datagrid AI in the side panel, changes made in one application reflect in the other. For example, if you modify an agent in the Datagrid standalone product, the new logic will apply when you use it in the Procore Datagrid AI side panel.

• Save Often: Ensure you click Save Changes at the top of your configuration editing panel.

• This tutorial helps you customize an agent, but Datagrid also has prompt controls affecting how the agent is used during a conversation. To refine your use of an agent, see optimizing AI results using prompt controls.

1. Open the agent Configuration panel for the agent you want to modify.

   • In Procore: Click themore menu next to the agent's name in a chat thread.

   • In Datagrid's standalone product: Click themore menu next to the agent's name in the left panel or from the chat thread when using it.

2. The Configuration panel is divided into specialized modules that affect how the agent thinks and what it can access. Explore your customization options:

Instructions

Define the "System Prompt" or persona. Tell the agent who it is (e.g., "You are a Senior Project Manager") and how it should respond (e.g., "Always provide answers in bullet points" ). This is the foundation of the agent's behavior.

System Instructions

This subsection dictates the Agent's baseline behavior and decision-making logic.

Why do this? This configuration tells the AI what document types or goals to prioritize, though a human must always verify outputs from AI.

1. Define the Role: Clearly state who the agent is (e.g., "You are a Deep Search agent specialized in fast, defensible answers" ).

2. Set the Primary Objective: Instruct the agent on its main goal, such as delivering fact-evidence and explicitly identifying conflicts.

3. Establish Operating Principles: Provide rules for data handling, such as using the most current revision and never resolving conflicts silently.

4. Set the Authoritative Hierarchy: Rank your document types (e.g., 1. Specs, 2. Drawings, 3. Addenda) to tell the AI which source to trust if information overlaps.

5. Define Boundaries: Set hard stops, such as instructing the agent to state clearly when information is missing.

6. Attach Reference Files: Click the Attach files to System Instructions button to upload a guide that the agent should always reference for its core logic.

Planning Strategy

This subsection defines the step-by-step cognitive process the agent follows to answer a query.

Why do this? This makes the agent more likely to perform a systematic search and verify loop. This can reduce errors and makes a more visible audit trail of how the AI arrived at its conclusion.

• Step 1 — Parse the question: Instruct the agent to first identify the affected scope (e.g., drawing sheet, system, product) and decision context (e.g., compliance, procurement).

• Step 2 — Targeted search/filter: Command the agent to extract only relevant text directly affecting the response and to capture exact section or sheet numbers.

• Step 3 — Cross-verification: Require the agent to check additional sources (like Addenda) to see if the initial search result was overridden.

Template Inheritance

This feature allows you to set high-level rules across multiple agents.

Why do this? Inheritance allows a company to set global and compliance standards once, while still giving project teams the flexibility to tweak an agent for a specific phase of work (like focusing purely on HVAC commissioning). This helps standardize agents across a group.

How to use it: Note the "This setting is inherited from the Agent Template" indicator. You can choose to keep these system-wide settings or click the Edit icon to override them for a specific, specialized task.

Tools

Customize the abilities of the agent. This allows the AI to perform specific tasks like searching the web, analyzing CSV files, or using Procore-specific search tools. Toggle tools on or off based on the agent's specific job.

Web Capabilities

This group allows the agent to interact with the broader internet to supplement project data.

Why do this? While your project files (grounding) are the priority, construction often requires verifying external data like current building codes, material lead times, or weather impacts. These tools can help bridge the gap between your project data and the real world.

• Toggle Web Search and Access Links to allow the agent to browse the live web and follow specific URLs provided in a chat.

• Toggle Company Research or People Research for background checks on vendors or partners.

• Use the NEC Calculator toggle for electrical code compliance checks.

Data Processing

Use these tools to transform raw document content into actionable information.

Why do this? Much of construction administration involves reformatting data (e.g., turning a field note into an RFI draft). These tools automate that "busy work," helping with consistency and saving time on manual entry.

• Toggle Structure Extracted Data and Classify Data to have the agent organize text into cleaner formats.

• Enable Document Generator and PDF Form Processing to allow the agent to draft or fill out official project documents.

• Toggle Run Code for advanced mathematical analysis or custom data manipulation.

Knowledge Management

These tools define how the Agent reads and understands your uploaded project attachments.

Why do this? Construction documents are uniquely complex. Semantic search helps the AI to understand, for example, that a query about HVAC cooling should also look at chiller specs, even if the words do not match exactly. This is often at the core of improving AI performance.

• Toggle Semantic Search and Analyze PDFs using AI to move beyond simple keyword matching.

• Toggle Detect Attachment and PDF Page Information to help the AI navigate large drawing sets.

• Toggle Memory to allow the agent to recall context from previous interactions within the same project.

Enhanced Responses

Control how the Agent presents its final answers to you.

Why do this? AI is most useful when its output is ready to use. Instead of reading a 500-word summary, a table allows for instant comparison, and a download link helps you share the findings with the field team.

• Toggle In-response Tables to tell the agent to present complex comparisons in a structured table format.

• Enable Provide Download Link to allow the agent to generate and offer a file for you to download directly from the chat.

Actions (Beta)

The 'Actions' section is where you enable your agent to interact with the external software tools your team uses every day.

Why do this? Connecting these actions transforms the agent into a more helpful team tool. Instead of just answering a question, it can perform multi-step workflows—for example, analyzing a site photo and then automatically drafting an RFI in Procore or notifying a subcontractor via Slack about a newly detected issue. This can remove some of your manual context switching between apps and help project data flow through your organization.

• Use the Search actions bar to locate a specific integration among the ones available

• Toggle communication tools like Slack or Microsoft Teams to 'On' to allow the agent to send summaries or notifications directly to project channels.

• Enable document storage actions such as SharePoint, Egnyte, Google Drive, or Box to let the agent save processed data or pull in additional context from external folders.

• Toggle construction-specific platforms like Procore Enterprise, Fieldwire, ACC (Autodesk Construction Cloud), or Trimble Connect to synchronize live project data.

• Enable administrative and legal tools like DocuSign or Notion to automate the drafting and signing of project documentation.

MCP Servers

The 'MCP Servers' section allows you to connect external tools and live data feeds to your agent through the Model Context Protocol (MCP).

Why do this? While 'Actions' allow the agent to talk to other apps, MCP servers allow the agent to securely read from, write to, and take action in your existing technical workflows all in one place. This helps connect live, proprietary data—like a real-time BIM database or a custom material procurement system—directly into the agent’s reasoning process.

Note that clicking MCP Servers in the agent configuration panel will currently show you which servers are available, but to add a new one, you must navigate to the global settings menu manually in the Datagrid standalone product. MCP servers often house secure enterprise-level information across a company so their scope is beyond the individual agent configuration level. Note: New MCP servers you add may not be accessible from the Datagrid AI in Procore.

1. Click the Settings gear icon at the bottom of the left-hand navigation bar of the Datagrid standalone product.

2. Under the 'Procore Teamspace' category, click MCP Servers.

3. Click Register MCP Server to open the connection dialog.

4. Provide a 'Display Name' (like 'Structural Database' or 'Global Search') so you can easily identify the server later.

5. Enter the 'Server URL' for the external service you want to connect.

6. Click Register Server to finalize the connection and make it available for use by your agents.

Knowledge

The 'Knowledge' section is an important part of your agent's setup, as it defines the specific documents, datasets, and live project data the agent uses to ground its answers. By curating this knowledge base, you help the agent provides accurate, project-specific information rather than general AI responses.

Configure the datasets and source files that your agent will use.

Teamspace Knowledge

This area manages the specific datasets available within your immediate project or team environment.

Why do this? This allows you to more precisely control which project information the agent is "aware" of. For example, you might want a specialized agent to only reference 'Procurement Schedule' data to avoid noise from unrelated documents.

1. Toggle onAllow access to all current and future datasets on if you want the agent to automatically include any new project data as it is added to the teamspace.
   

   Pro-tip

   Toggling OFF Allow access to all current and future dataset prevents the AI from cross-contaminating info from other user's datasets. Depending on how many users are in your teamspace, this prevents your agent from ingesting data that is irrelevant or conflicting with your goals.

2. Use the Search Knowledge bar to find specific datasets by name.

3. Individually select or deselect datasets using the checkboxes next to items like 'New Procore Dataset' or 'Procurement Schedule Analysis & Actions'.

4. Use Select All or None for quick bulk selection.

5. Click Refresh list to ensure you are seeing the most recent data additions.

Organization Knowledge

This section allows your agent to pull from broader, company-wide data sources.

Why do this? Using organizational knowledge allows your agent to leverage lessons learned and standard specifications from previous projects, so your project may benefit from company-wide expertise.

1. Browse through shared organization datasets.

2. Click Add Dataset to bring in new organizational data.

Adding New Knowledge Sources

You can bring in data from local files or external software products.

Why do this? Importing diverse data sources—like syncing your Primavera P6 schedule alongside your Procore drawings—allows the agent to perform complex, cross-document reasoning. This helps the agent answer high-level questions like, Are the current lead times in my inventory going to delay the concrete pour scheduled in my P6 plan?.

• Click Add Dataset and then select Upload File to add local PDFs, spreadsheets, or technical drawings directly to the agent's memory.

• Click Add Dataset and select Connect Apps to open the 'Pick an Import Source' dialog.

• Choose from categories like 'Construction', 'Finance', or 'Web Data' to filter your sources.

• Select specific platforms such as Procore, Autodesk Construction Cloud (ACC), or P6 Primavera to import live data feeds.

• Use cloud storage connections like Google Drive, Box, OneDrive, or Google Sheets to sync project folders.

Model

The 'Model' section allows you to configure the specific artificial intelligence models and behavioral settings that alter how your agent thinks, speaks, and interacts with project data.

Agent Model

Choose the version of the Datagrid agent brain that acts as the primary orchestrator for your project tasks.

Why do this? The agent brain is the coordinator. Selecting a more advanced brain like Magpie-2.0 may help the agent navigate complex cross-document logic—such as identifying discrepancies between documents.

• Click the Disinherit gear icon next to the 'Agent Model' title if the setting is currently locked by a template; this will reveal the available choices.

• Open the dropdown and select Magpie-1.1-flash for quick, simple semantic answers.

• Choose Magpie-2.0 for more complex, multi-step reasoning tasks that require improved performance and enhanced capabilities.

LLM Response Generation

Pick the underlying Large Language Model (LLM) that will generate the final text responses in your chat.

Why do this? LLMs have unique strengths. Gemini models are often preferred for massive document sets (like a 2,000-page project manual) due to their large context windows, while GPT-4o might be selected for more precise logic in routing complex technical queries, and Claude 4.6 when more precise synthesis and nuanced adherence to complex formatting is needed.

1. Click the Disinherit gear icon to unlock the LLM selection.

2. Scroll through the dropdown to select your preferred model. Select specialized models if your project requires heavy professional-grade image analysis or generation.

Model behavior

Fine-tune the agent’s creativity, faithfulness to the data, and word choice.

Why do this? Precision is more important than creative writing for some tasks. Keeping the 'Temperature' lower for those tasks ensures the agent stays grounded, reporting only verified facts from the drawings and specs rather than imagining new details to sound more helpful.

• Move the Temperature slider to control randomness; lower values like 0.1 make the agent more focused and predictable.

• Adjust the Top P slider to refine how the model selects words based on probability.

• Note that for specific models like claude-opus-4.6, you should only edit one of these sliders, as the model typically supports either Temperature or Top P, but not both simultaneously.

Voice Preset' and 'Language

Configure the auditory personality and language for voice-based interactions.

Why do this? If you are using the agent in the field via voice commands, choosing a firm, clear voice like Nova or Sage may make it easier to hear and understand the AI's technical guidance over the noise of a busy jobsite.

• Open the 'Voice Preset' dropdown and scroll to select a vocal tone, such as Spark (bright and higher pitch), Sage (informative and lower pitch), or Nova (firm and middle pitch).

• Set the Language to your local code, such as en-US, to ensure the agent understands and communicates with the correct regional context.
  Note: If you cannot select a new language, the current one may be the only currently-available language.

Mobile Access

The 'Mobile Access' section is where you enable your agent to communicate with your team in the field through SMS via a dedicated, unique phone number.

Why do this? Since field workers don't always have a desktop open or capability of to navigate the full Procore mobile app, SMS provides a lower-friction, universal interface. By texting the agent, field teams can ask the AI for technical data—like asking for a specific material lead time or a clarification from a spec—while remaining more hands-on with their work. This helps make the agent's intelligence accessible even in areas with limited data connectivity, where SMS is likely more reliable.

Provision a dedicated phone number for your agent to enable SMS-based field communication:

1. Select the correct Country from the dropdown menu (currently limited to 'United States' and 'Canada').

2. Click the Acquire phone number button to initiate the assignment process.

3. Review the 'Acquire AI Agent Phone Number' confirmation dialog, which notes that setting up the number and sending individual messages will consume your project's allocated credits.

4. Click Acquire to finalize the setup and receive the unique number assigned to your agent.

5. Share this unique number with your 'selected contacts' (field leads, subcontractors, etc.) so they can begin messaging the agent directly.

Chat via Microsoft Teams

Integrate the agent directly into your company’s Teams environment. Once configured, the agent can be "tagged" in a Teams channel to provide project insights during a group chat.

Why do this? Integrating with Teams brings AI directly into the room where decisions are being made. This allows multiple team members to see the agent's response simultaneously, helping everyone work from the same information. Note that for privacy and security, the agent does not read any channel conversations unless a user specifically tags it, keeping your internal team discussions private.

1. Navigate to the 'Chat via Microsoft Teams' configuration screen.

2. Click Connect to Microsoft Teams to begin the authorization process.

3. Follow the prompts to sign in to your Microsoft account and grant the necessary permissions for the agent to access your workspace.

4. Once connected, use the Create Connector dialog to link the agent to specific channels or enable direct messaging.

5. Click Add Connector to finalize the link between the agent and your selected Teams channel.

6. To interact with the agent in a channel, simply tag @Datagrid followed by your question.

Template Inheritance

The 'Template Inheritance' section allows you to standardize initial agent configurations across multiple agents by linking them to a master template. This supports consistency in instructions, tools, and project knowledge while still allowing for project-specific overrides when necessary. You can view which master template this agent is based on and revert changes or automatically apply global updates from a parent template.

Why do this? Template inheritance can help scale AI across a large construction firm. It can reduce the need to manually re-configure complex logic for every new jobsite, saving time and reducing the risk of a field agent providing unverified or non-compliant information.

Managing Agent Templates

Follow these steps to connect or disconnect your agent from an organizational standard.

1. Open Settings: Click Template Inheritance in the agent configuration panel.

2. Link to a Standard: Select Inherit from Template and choose a configuration (e.g., [TG] Deep Search) from the dropdown.

3. Customize/Detach: To make unique changes, click Detach Template. This breaks the link while keeping your current settings as a baseline.

4. Re-Sync: To revert a custom setting back to the template, click the dotted Inherit icon and select Apply Template.

Identifying Inheritance Status

Look for the Inherit (gear-head) icon next to sections like 'Tools' or 'Actions' to see if you are in sync.

• Blue icon: The setting is currently synced with the master template.

• Dotted outline icon: The setting has been customized for this specific agent.

Organization-Level Management

Requires an Enterprise-tier subscription and access to the Datagrid Standalone hub.

• Create New Standards: Once an agent is detached, click Create Template from this Agent to save it as a new reusable standard for the company.

• Global Updates: Navigate to Settings > Organization Settings > Manage Templates. Click Configure to edit master instructions and tools for all agents using that template.