A shared AI engine that powers 21 industry-specific products. Multi-agent architecture, omnichannel communications, and industry compliance built in from day one.
Barrett Agentic is a multi-tenant platform where each vertical gets its own set of specialized AI agents, compliance rules, workflows, and billing configuration. The core engine is shared; the industry expertise is unique to each vertical.
Each vertical runs multiple specialized AI agents rather than one monolithic model. This lets each agent be an expert in its narrow domain while the supervisor coordinates their work.
The Supervisor agent receives every inbound message and determines which specialist should handle it. If a billing question comes in during an intake conversation, the Supervisor hands off to the Billing Agent with full context. Agents can also run in parallel for independent tasks like sending a reminder while processing a payment.
Every agent runs on Anthropic's Claude (claude-sonnet-4-20250514 by default, configurable per agent). We chose Claude for its reliability, instruction-following precision, and ability to handle nuanced industry conversations without hallucination.
Each vertical includes a document ingestion pipeline that converts business documents (SOPs, rate sheets, compliance guides) into vector embeddings stored in pgvector. When an agent needs to answer a question, it performs hybrid search (vector similarity + BM25 keyword matching) to retrieve the most relevant context before generating a response.
Agents maintain conversation history per client, with automatic summarization for long threads. This means a client can text "same as last time" and the agent knows exactly what that means because it has the full relationship context.
Every vertical supports three communication channels out of the box, with no additional configuration required.
| Channel | Provider | Capabilities |
|---|---|---|
| SMS | Twilio | Two-way texting, automated reminders, appointment confirmations, status updates |
| Resend | Transactional email, drip sequences, document delivery, inbound email processing | |
| Voice | Twilio | Inbound IVR, call routing, voicemail transcription, outbound notifications |
| Chat | WebSocket | Real-time web chat, dashboard messaging, internal team communication |
Primary database with row-level security. All data isolated by organization. pgvector extension for embedding storage and similarity search.
Async API backend with 70+ endpoints. Handles webhooks, agent orchestration, billing, and real-time WebSocket connections.
Production deployment on Railway with auto-scaling. GitHub-linked CI/CD pipeline for zero-downtime deployments.
Subscription billing with webhook-driven lifecycle. Three pricing tiers per vertical, free trials, and self-serve checkout.
DNS, CDN, and static page hosting for all 21 vertical landing sites. Workers for edge logic and routing.
Primary AI model for all agent reasoning. Configurable model selection per agent for cost/performance optimization.
Every vertical has industry-specific compliance rules defined in its YAML configuration. These rules are enforced at the agent level, meaning agents cannot take actions that violate regulations.
Compliance rules are versioned and updated as regulations change. The platform maintains a full audit trail of every agent action, every message sent, and every decision made.
Each vertical is defined by a YAML configuration file that specifies its agents, compliance rules, workflows, integrations, pricing tiers, and brand identity. Adding a new vertical means writing one YAML file and creating industry-specific agent personas. The platform handles everything else.
This architecture means we can launch a new vertical in days, not months. The shared engine handles authentication, billing, communications, deployment, and monitoring. The YAML config handles the industry-specific intelligence.