Cadence

Cadence is an open-source workflow engine for developers building fault-tolerant, stateful distributed applications. It suits application developers who write long-running workflows and platform engineers running orchestration as shared infrastructure. Its strongest case is durable execution without per-run charges; the trade-off is operating it yourself and writing replay-safe, deterministic code.

Overview

Cadence turns workflow logic into durable functions that can resume after process restarts, infrastructure failures, or deliberate pauses. That makes it a serious foundation for stateful background processes rather than a simple task list. Event triggers, automatic retries, execution observability, and human approvals cover the control points many production workflows need.

The project dates to 2017 and is open source. You can deploy it with embedded SQLite, Docker Compose, or Kubernetes through Helm. A managed Uber cluster is also described for Uber internal teams, but Cadence's general deployment model remains self-hosted. Readers comparing categories can place it alongside Process Orchestration Software and Workflow Orchestration Software.

Key features

Durable, replayable workflows

Cadence reconstructs workflow state by replaying code, so workflow code must be deterministic. This is a meaningful engineering constraint: teams gain recovery across failures, but ordinary nondeterministic behavior can break replay and demands discipline from developers.

Scheduling and operations

Schedules handle recurring runs, overlap policies, pausing, updates, backfills, and execution history. Those controls suit recurring data, maintenance, and business processes where operators need to revise timing without losing a record of prior executions.

Integrations and protection

Official SDKs cover Go, Java, and Python, with a community Ruby client. Production workflow history can use Cassandra, MySQL, or PostgreSQL; visibility can use Elasticsearch, OpenSearch, or Apache Pinot. Kafka supports search indexing, while Prometheus, StatsD, and M3 provide metrics. TLS is configurable for backing-service connections, and frontend SDK connections support mutual TLS.

A custom data converter can compress or encrypt payloads, or replace large payloads with references to external object storage. This gives platform teams practical control over sensitive or oversized workflow data, though implementation remains their responsibility. Community support runs through Stack Overflow, CNCF Slack, GitHub Issues, and contributors rather than a stated commercial support tier.

Pricing

The free plan is the whole product's open-source distribution, not a capped starter edition. There are no seat quotas, execution allowances, or per-run charges published for it, making cost predictable at scale. The cost shifts to your own infrastructure, deployment, storage, security, and operations work because the model is self-hosted.

Platforms

Cadence is offered through an API, web interface, Linux deployment, and self-hosted operation. Kubernetes with Helm and Docker Compose cover common infrastructure paths, while embedded SQLite offers a lighter deployment option. The listed SDK languages make it most natural for teams already building services in Go, Java, or Python; Ruby support comes from the community.

Who it's for

Choose Cadence when your engineering team needs durable, stateful workflows that survive failures and pauses, and when it can operate the underlying service and databases. It is particularly suitable as shared infrastructure for platform teams supporting multiple application workflows. It is a weaker fit for organizations seeking a hosted, low-operations workflow product or for developers unwilling to design around deterministic replay.

Pros and cons

Pros

• Durable functions resume after process and infrastructure failures, preserving long-running work.
• Scheduling includes overlap rules, backfills, pausing, updates, and execution history instead of only fixed recurring triggers.
• No license key, usage limits, or per-execution fees make the open-source plan suitable for high-volume workloads.
• Broad storage, visibility, messaging, metrics, and payload-protection integrations support production architectures.
• Mutual TLS and configurable TLS address transport security between SDKs and services.

Cons

• Self-hosting leaves deployment, databases, upgrades, and operational reliability with your team.
• Replay requires deterministic workflow code, imposing rules that can complicate application design.
• Ruby has only a community client, unlike the official Go, Java, and Python SDKs.
• Support is community-based, with no stated paid support plan.

Alternatives

Joget is a freemium, multi-platform option if you need a free three-user tier and a broader low-code application focus. Bonita suits teams wanting a free, community-supported plan capped at 150 process instances per month. Camunda is worth choosing when a self-managed development plan and a freemium path are preferable.

Dapr fits developers seeking free durable workflows with native integrations and workflow visualizers for development and non-critical workloads. Bizagi Modeler is a simpler modeling choice, with a free one-user plan and local storage. Cloudflare Workflows is the alternative for API and web workloads using Workers Free, with shared request and CPU limits.

Hatchet offers a hosted-style pay-as-you-go route with 1 million free runs and events per month. Inngest is a freemium managed option with a five-seat, 50,000-execution free plan and a $99 a month Pro starting tier.

Verdict

Cadence is a strong choice for engineering-led organizations that need durable, replay-safe workflows and are prepared to run orchestration infrastructure themselves. Its open-source economics, recovery model, scheduling controls, and integration breadth are compelling. Look elsewhere if you want a hosted service, minimal operational ownership, or workflow code that cannot reliably satisfy deterministic replay.