Operations Guide¶

FastAgentic provides a unified operations layer for AI agents. This guide covers deployment, configuration, observability, security, and operational procedures.

Why DevOps Teams Choose FastAgentic¶

The Problem: AI Agent Deployment Chaos¶

Without FastAgentic, deploying AI agents means: - Each agent framework has different deployment patterns - Custom instrumentation required for observability - No unified governance across agent types - Compliance gaps (audit trails, cost tracking) - Operational complexity multiplies with each new framework

The Solution: Unified Agent Operations¶

1. Deploy Once, Run Any Framework

# Same deployment for PydanticAI, LangGraph, CrewAI, LangChain
docker run -e FASTAGENTIC_OIDC_ISSUER=https://auth.company.com \
           -e FASTAGENTIC_DURABLE_STORE=postgres://... \
           fastagentic/runtime:latest

2. Built-in Governance Without Custom Code - Cost tracking per run, user, tenant (out of the box) - Audit trails satisfy SOC2/GDPR requirements - Policy enforcement (quotas, rate limits) at runtime - Role-based access with OAuth2/OIDC integration

3. Production Reliability by Default - Checkpoint durability (Redis/Postgres/S3) - Automatic resume on failure - Idempotent execution with run IDs - Graceful degradation under load

4. Zero-Instrumentation Observability - OpenTelemetry traces span all agent operations - Standard RED metrics per endpoint - Token usage and cost metrics - Correlated logs with trace IDs

Quick Production Deployment¶

Docker¶

docker run -d \
  --name fastagentic \
  -p 8000:8000 \
  -e FASTAGENTIC_OIDC_ISSUER=https://auth.company.com \
  -e FASTAGENTIC_DURABLE_STORE=redis://redis:6379 \
  -e FASTAGENTIC_OTEL_ENDPOINT=http://otel-collector:4318 \
  fastagentic/runtime:latest

Kubernetes¶

helm repo add fastagentic https://github.com/neul-labs/fastagentic/tree/main/charts
helm install my-agents fastagentic/runtime \
  --set auth.oidcIssuer=https://auth.company.com \
  --set durability.backend=postgres \
  --set durability.connectionString=postgresql://... \
  --set telemetry.enabled=true

Operational Architecture¶

                    ┌─────────────────┐
                    │  Load Balancer  │
                    │  (TLS, routing) │
                    └────────┬────────┘
                             │
              ┌──────────────┼──────────────┐
              │              │              │
              ▼              ▼              ▼
        ┌──────────┐  ┌──────────┐  ┌──────────┐
        │FastAgentic│  │FastAgentic│  │FastAgentic│
        │ Instance 1│  │ Instance 2│  │ Instance N│
        └─────┬─────┘  └─────┬─────┘  └─────┬─────┘
              │              │              │
              └──────────────┼──────────────┘
                             │
         ┌───────────────────┼───────────────────┐
         │                   │                   │
         ▼                   ▼                   ▼
   ┌──────────┐        ┌──────────┐       ┌──────────┐
   │  Durable │        │   OTEL   │       │   LLM    │
   │  Store   │        │ Collector│       │ Providers│
   │(Postgres)│        │          │       │          │
   └──────────┘        └──────────┘       └──────────┘

Documentation Sections¶

Deployment¶

Docker Deployment - Container deployment guide
Kubernetes Deployment - Production K8s with Helm
Serverless Deployment - AWS Lambda, Cloud Run patterns

Configuration¶

Environment Variables - Complete reference
Secrets Management - Secure credential handling

Observability¶

Metrics Reference - Available metrics and dashboards
Distributed Tracing - OpenTelemetry setup
Alerting Guide - Alert rules and runbooks

Security¶

Security Hardening - Production security guide
Compliance Guide - SOC2, GDPR, audit trails

Runbooks¶

Troubleshooting - Common issues and solutions

Key Operational Features¶

Health Checks¶

FastAgentic exposes standard health endpoints:

Endpoint	Purpose	Returns
`/health/live`	Liveness probe	`200` if process alive
`/health/ready`	Readiness probe	`200` if accepting traffic
`/health/startup`	Startup probe	`200` when initialized

Graceful Shutdown¶

On SIGTERM: 1. Stop accepting new requests 2. Wait for in-flight runs to checkpoint 3. Drain background workers 4. Close database connections 5. Exit cleanly

Configure drain timeout: FASTAGENTIC_SHUTDOWN_TIMEOUT=30

Multi-Tenancy¶

FastAgentic supports tenant isolation:

# config/settings.yaml
tenancy:
  enabled: true
  isolation: database  # or "schema" or "row"
  header: X-Tenant-ID

Each tenant gets: - Separate durable stores (or schemas) - Independent quotas and rate limits - Isolated audit trails - Per-tenant cost tracking

Horizontal Scaling¶

FastAgentic instances are stateless. Scale horizontally behind a load balancer:

Sticky sessions not required
All state in durable store (Redis/Postgres)
Background workers share centralized queue
No instance-to-instance communication needed

Production Checklist¶

Before going to production:

Next Steps¶

Docker Deployment - Start with containers
Environment Variables - Configuration reference
Metrics Reference - Set up dashboards