Connecting AI Agents

BuildChart uses a BYOK (Bring Your Own Key) model for AI agents. You provide your own AI agent and LLM API key — BuildChart provides the data and execution layer. This means you stay in control of which AI model you use, what data it sees, and how much you spend on inference.

Three Connection Methods

There are three ways to connect an AI agent to BuildChart, depending on your architecture and use case:

1. REST API — Your agent polls or reacts to data via standard HTTP endpoints. This is the most flexible approach and works with any language or framework. The agent makes authenticated requests using an API key to read project data, update task statuses, or create draft actions.
2. Webhooks — BuildChart pushes events to your agent's endpoint in real time. When a task becomes overdue or a status changes, your agent receives an instant notification and can respond immediately without polling.
3. MCP Server — Connect Claude or any MCP-compatible agent directly to BuildChart tools. The MCP (Model Context Protocol) server exposes BuildChart operations as tools that the agent can call natively, making the integration seamless and conversational.

The Two-Phase Action Pattern

For high-risk actions — sending emails, sharing files, making financial commitments — BuildChart enforces a two-phase approval pattern:

1. The agent creates a draft action describing what it wants to do (e.g., "Send RFQ email to Smith Plumbing").
2. You review and approve or reject the action in the BuildChart app. You see exactly what the agent intends to do before anything happens.
3. Once approved, BuildChart executes the action on behalf of the agent.

Low-risk actions like status updates and internal notifications can be configured to auto-execute on the Business tier, skipping the approval step for trusted workflows.

What Agents Can Do

• Read project data, task lists, schedules, and activity logs.
• Update task statuses and percent complete.
• Create draft communications (emails, notifications) for your review.
• Analyze schedules and detect overdue or at-risk tasks.
• Generate reports and summaries from project data.

What Agents Cannot Do

• Send external emails without your explicit approval.
• Share files with external parties without approval.
• Make financial commitments or authorize payments.
• Bypass the approval workflow for high-risk actions.

These guardrails ensure that your agent is a powerful assistant without ever taking irreversible action on your behalf without consent.