Features

Everything you need
to run AI in production.

PlexRun is an opinionated runtime — not a collection of utilities. Every feature is designed to work together so your team can ship reliable AI workflows without reinventing the infrastructure.

Request early access

Workflow Engine

Core

DAG-based execution with native support for branching, fan-out, fan-in, and conditional routing. Define complex pipelines in Python or YAML without managing execution state manually.

Directed acyclic graph (DAG) model for pipeline definition
Conditional branching based on step outputs
Fan-out to N parallel branches, fan-in with aggregation
Nested sub-workflows for modular composition

Multi-Agent Orchestration

Core

Run coordinated agent swarms with shared state, message passing, and supervised execution. PlexRun handles agent lifecycle, inter-agent communication, and failure isolation.

Shared state store across agents in a workflow
Typed message passing between agent steps
Supervisor patterns with automatic re-spawning
Human-in-the-loop step types for approval flows

Distributed Execution

Infrastructure

Auto-scaling worker pools across regions. Sub-second cold starts on serverless runtime. Run thousands of concurrent workflows without managing infrastructure.

Serverless-first with sub-second cold starts
Dedicated worker pools for Enterprise
Multi-region deployment (us-east-1, eu-west-1, ap-southeast-1)
Auto-scaling based on queue depth

Real-Time Observability

Observability

Per-step traces, token usage, latency, and cost. Replay any execution from any point. Export spans and metrics to your existing observability stack via OpenTelemetry.

Per-step execution traces with timestamps
Token usage and cost attribution per step
Full execution replay from any checkpoint
OpenTelemetry export for spans, metrics, and logs

Retries & Idempotency

Reliability

Exponential backoff, circuit breakers, dead-letter queues, and idempotent step keys built in. Never worry about duplicate execution or cascading failures again.

Configurable retry policies per step
Exponential backoff with jitter
Circuit breaker patterns for downstream services
Dead-letter queues for unrecoverable failures

Event-Driven Triggers

Integrations

Trigger workflows from webhooks, cron schedules, queue messages, or API calls. React to events from any upstream system in real time.

Webhook triggers with signature verification
Cron and interval-based scheduling
SQS / Kafka message queue integration
REST API trigger with sync or async response

Versioning & Rollback

Developer

Every deployed workflow is versioned. Compare revisions, deploy to staging before production, and roll back to any previous version in one command.

Immutable versioned workflow deployments
Side-by-side revision comparison
Staged rollouts (canary, blue/green)
One-command rollback via CLI or API

Enterprise Security

Security

SOC 2 in progress. TLS in transit, AES-256 at rest. RBAC, VPC peering, and SSO for Enterprise. Your data never leaves your cloud account.

Role-based access control (RBAC)
VPC peering and PrivateLink for Enterprise
SSO / SAML 2.0 support
Audit logs with tamper-evident storage

API-First Architecture