Platform Architecture

Built for enterprise teams that need production-grade infrastructure, not prototypes.

Agent Runtime

AgentaWorks agents run as isolated processes with configurable model backends, session management, and state persistence. Each agent maintains its own context, memory, and skill registry.

  • •Multi-model support: Claude, GPT-4, Gemini, and custom endpoints
  • •Session management: Persistent conversations with configurable timeouts
  • •Graceful scaling: Horizontal scaling with load balancing across replicas

Skill System

Skills are discrete capabilities that extend what an agent can do. They're registered via API and invoked by the agent runtime based on conversation context.

REST Integrations

Call external APIs with auth, retry, and error handling

Data Lookups

Query databases, vector stores, or search indexes

Channel Adapters

Send messages, notifications, or trigger workflows

Channels

Agents communicate through channels — interfaces that connect agents to users, systems, or other agents.

  • ✓Telegram: Direct messages and group integration with button support
  • ✓REST Endpoint: Synchronous HTTP API for programmatic access
  • ○WebChat: Embeddable widget for customer-facing deployments (coming soon)

Deployment

Start with Docker Compose on a single machine, scale to Kubernetes when you need orchestration. The same config works everywhere.

Development & Testing

  • • Docker Compose for local development
  • • Hot reload for rapid iteration
  • • SQLite or PostgreSQL for data

Production (On-Prem)

  • • Kubernetes manifests included
  • • Multi-replica deployments
  • • Your infrastructure, your rules

For Tier 3 clients with air-gapped environments: full on-premise installation with no external dependencies.

Security

Authentication & Authorization

  • • JWT-based authentication
  • • RBAC with per-agent permissions
  • • Org-level and user-level roles
  • • SAML/SSO integration (Phase 2)

Audit & Compliance

  • • Append-only audit log
  • • Every API call recorded
  • • Agent conversation history
  • • Exportable for compliance review

Status & Telemetry

Real-time visibility into agent health, performance, and activity across your organization.

  • •10-second polling: REST endpoint for current agent status
  • •SSE stream: Server-sent events for real-time updates
  • •Org status board: Dashboard showing all agents, deployments, and activity

API in Action

# Create an agent
POST /api/v1/agents
{
  "name": "Epicor Assistant",
  "slug": "epicor-assistant",
  "model": "claude-sonnet-4-5",
  "system_prompt": "You are an expert in CW Epicor...",
  "skills": ["epicor-lookup", "inventory-check"]
}

# Deploy it
POST /api/v1/deployments
{
  "agent_id": "...",
  "replicas": 1
}

# Check status
GET /api/v1/status/epicor-assistant
{
  "status": "running",
  "replicas": 1,
  "uptime": 3600,
  "last_activity": "2026-03-09T08:45:12Z"
}