The core idea
A PathPilot agent is a template plus configuration, executed inside an organization-scoped runtime with access to your data and tools. You configure agents once, then run them against real customer interactions across voice, chat, and email.Building blocks
Templates
Reusable blueprints that define an agent’s prompt, tools, and workflow. Instances are created from a template and customized per use case.
Knowledge Bases
Vector stores that ground agent responses in your policies, playbooks, and documents. Content comes from URL crawls or file uploads.
Workflows
LangGraph-based orchestration for multi-step tasks like QA scoring, KYC review, or campaign runs. Executed in test or production mode.
Tools & Integrations
Agents call typed tools (knowledge search, ticket lookups, decisioning engines, Twilio, ElevenLabs) to take real action.
How execution works
1
An event arrives
A customer sends a WhatsApp message, a call is dispatched, or a workflow is triggered on a schedule.
2
The runtime hydrates context
The right template is loaded, configuration is resolved (including secrets), and the agent is scoped to the calling organization.
3
The agent works
It retrieves from the knowledge base if needed, calls tools, follows its workflow, and generates a response or action.
4
Results are recorded
Executions, transcripts, and outcomes are stored so you can review quality, audit decisions, and improve prompts and rubrics.
Multi-tenant and auditable
Everything runs per organization. Data, credentials, and agent instances are isolated. Every execution is logged and linkable back to the campaign, ticket, or application it belongs to, so you can audit decisions and coach agents with confidence.Deeper technical documentation (template schemas, workflow node contracts, runtime internals) is available to customers after access is granted. Request access at getpathpilot.com.