How Sheptra compares

Most of the executable core: every gateway but complex, timer / message / signal / error / escalation / compensation / link / terminate events, event sub-processes, non-interrupting boundary events, call activities, multi-instance. Transactions, cancel and conditional events, complex gateways, ad-hoc sub-processes and standard loops are rejected at deploy time with the element named, never mid-instance.

Most of the executable set, including event sub-processes, multi-instance, compensation, escalation, conditional events and ad-hoc sub-processes. Not executable: transactions and cancel events, link events, complex gateways, multiple events and the standard loop marker.

Timer, signal, message, conditional, error and terminate events; exclusive, inclusive, parallel and event-based gateways; form, manual and script tasks; sub-processes. Not documented: transactions, cancel and link events, complex gateways, compensation, escalation, event sub-processes, multi-instance markers.

Decision tables with all seven hit policies, literal expressions and required decisions, evaluated inside the engine by the business rule task — no job, one step. A deployed decision can also be evaluated directly as a dry run. Measured against the DMN TCK.

Decision tables with Unique, Any, First, Rule Order and Collect (sum, min, max, count) hit policies, evaluated by the DMN engine inside Zeebe when a business rule task calls them.

Decision Tables saved as DMN with First, Unique, Any, Collect (sum, min, max, count) and Rule Order hit policies, evaluated by the Drools engine. The docs do not say whether evaluation is embedded or a separate service.

Own interpreter: paths, arithmetic, comparisons, if / then / else, lists with filters, contexts, for / some / every, ranges and unary tests, ranges as values with the range builtins (before, after, overlaps, includes …), names containing spaces (as variables and as path segments, order.total price), named arguments, function literals with sort by a comparator (sort(orders, function(a, b) a.amount > b.amount)), dates, times and durations with their arithmetic and ordering (min / max / sort included), zoned times with an offset or an IANA zone id, and the string, number, list and calendar builtins. Not yet: sum / mean over durations. Conformance is measured against a shared corpus.

FEEL-Scala evaluates every BPMN and DMN expression in-engine, following the OMG FEEL specification; feelin covers Forms in the browser. No conformance-suite figure is published.

A FEEL-style syntax for gateway conditions, screen visibility and assignment rules, described as extending Symfony’s operators. No implementation is named and no claim of full FEEL is made.

FEEL only, evaluated in-engine and sandboxed by construction. Any other language runs in a worker you operate; no user code runs inside the engine, by design.

In-engine script tasks run FEEL only; Groovy, JavaScript, Python and the rest run through a job worker, such as the community script worker the docs point to.

Scripts in PHP, JavaScript, Python, Lua, C# and Java, each run in a per-language Docker script executor bundling the ProcessMaker SDK.

A form builder on form-js in the Modeler; a form deploys to an environment as a versioned engine resource, a user task names it by key, and Drive renders the version pinned on each task and completes the task through it. Not yet: file upload in a form, and public or start forms.

Camunda Forms built in the Web or Desktop Modeler, rendered by Tasklist or embedded through the open-source form-js library.

Screen Builder: drag-and-drop Form, Display, Email and Conversational screens, rendered in Form Tasks as a Request runs. Professional and Enterprise plans.

Browser modeler on bpmn-js with autosave, in-browser lint mirroring the engine’s deploy validators, token simulation, numbered versions with visual and XML diffs, restore, and deploy to a chosen environment and partition. DMN decisions and forms are edited in the same library.

Web Modeler on SaaS and Self-Managed, with named versions, visual and XML diffs and restore. On SaaS the deploy dialog picks the target cluster; on Self-Managed the cluster is preconfigured.

Browser Process Modeler with a version history, named versions and a “copy to latest” restore; DevLink links instances and moves asset bundles from development to production.

Not offered. Integrations are written as workers against the job API. A connector marketplace is not on the near-term roadmap.

An out-of-the-box catalogue — REST, SOAP, GraphQL, AWS, Azure, GCP, Slack, SQL and more — plus a Marketplace of partner and community connectors that are not commercially supported.

Eight documented connectors (email, DocuSign, Slack, PDF, IDP, decision task and a generic Data Connector for REST and Collections). No marketplace.

Long-poll activation across every job type in one request, complete / fail / throw-error, lease extension for long handlers, at-least-once delivery with an idempotency key on every write, a traceparent on every job. Plain HTTP; no SDK required.

Job workers activate jobs by type with long polling, then complete or fail them with a retry count. A job whose activation times out is handed to another worker without spending a retry.

No polling worker model. Script Tasks run in the executors, Data Connectors call external REST APIs (with an optional asynchronous callback), and outside code drives Requests through the REST API.

The retry budget is declared on the task in the model; a worker failing a job spends it, and exhausting it raises an incident an operator resolves one at a time or by a bulk retry.

Retries declared on the task definition in the model (default three); an incident is raised when they reach zero and is resolved from Operate or the API.

Retry attempts and wait time set per Script Task and Data Connector; when exhausted the Request enters Error status, and a super admin can retry it.

An alteration plan migrates a filtered set of running instances to another version, with an element-id mapping for renames and a preview that reports the affected count before anything changes. Applied by the scheduler under the instance lock, audited as the plan’s creator.

Migration through the REST or gRPC API, or in batch from Operate, mapping active elements onto the target version. An element cannot change type or nesting on the way.

Version changes are not reflected in Requests already in progress or completed; a new version applies to new Requests only.

Cancel tokens, activate elements or flows, and merge variables in one validated, atomic batch — from the instance diagram in Drive or the API — and as a bulk alteration across many instances. Every batch is audited with its instructions.

Activate, terminate and move element instances and set variables, from the API or Operate. Start events, boundary events and sequence flows cannot be targeted.

An errored Request can be retried, rolled back to an earlier Form Task, or completed by hand; an in-progress Request can be cancelled. No token move or added activity.

Message correlation by key with buffering and a replayable dead letter, broadcast signals, timer start / catch / boundary events, escalation, error propagation with worker-thrown errors, compensation when the model throws it, link events.

Messages with a correlation key, time-to-live and message id; broadcast signals; ISO 8601 and cron timers; escalation and compensation handlers — all in-engine.

Message events matched by name between pools, broadcast signals, timers, conditional and error events. No compensation or escalation events.

Everything under /v1 with a written policy: the surface only grows within a contract number, breaking changes ship under a new one, and every host reports the contract numbers it serves. The deprecation policy is published beside it.

A minor every April and October with eighteen months of standard maintenance and monthly alphas; the REST API is versioned as a whole (v2, semantic versioning); deprecations are listed in each release announcement.

Seasonal releases plus numbered updates — fourteen so far in the 2026 line — and a recommended thirty-day development-before-production window. No published support window, deprecation policy or API version policy.

REST with an OpenAPI document and an interactive reference on every environment, on both hosts. Sheptra.Client, a .NET worker SDK and typed client, is written against that contract but not yet published to NuGet. Plain HTTP is the contract underneath.

Orchestration Cluster REST API v2 with a downloadable OpenAPI document, plus gRPC. Java client, Spring Boot starter, TypeScript and Python SDKs, a C# SDK in technical preview, and a CLI.

An OpenAPI-compliant REST API with Swagger UI on every instance. The SDK is what scripts call from inside an executor (PHP, Lua, JavaScript, C#, Java, Python); no standalone client catalogue is published.

GET /v1/system reports the release, the contract numbers and the engine’s capability catalogue — one stable, additive-only identifier per feature — so a client asks a cell what it runs instead of inferring it from a version number.

The topology endpoint reports brokers, partitions, cluster size, replication factor, versions and partition health — the shape of the cluster, not the features it runs.

No capability or feature endpoint in the API documentation or the 2026 release notes.

Drive: instance search, a timeline interleaving engine steps and operator commands, incidents grouped by cause with one-click resolve, live variable edits, message publishing to unblock waits, modification from the diagram, alteration plans with progress — and a Tasks page to claim, release and complete user tasks.

Operate: instances, incidents, batch retry and cancel, modification and migration. Tasklist: the task inbox, rendering Camunda Forms.

Requests and Cases views with retry, rollback and cancel; a Tasks inbox with reassignment, priority, drafts and quick fill.

Organizations isolated by row-level security on a shared cell; each organization’s environments are scoped apart the same way, with optional sub-tenant partitions inside one. A dedicated database, dedicated compute, a deployment into your own cloud account and a choice of region are on the roadmap, labelled as such.

Tenant identifiers partition one cluster (Self-Managed, and SaaS on 8.8+ clusters); SaaS clusters are dedicated per customer. Seven GCP and six AWS regions, custom regions for Enterprise.

One instance per customer on AWS or Azure, with the hosting location chosen by the customer. No tenant-identifier model is documented; the GDPR statement names US data centres.

Per-workspace retention windows for execution history and finished instances, archived before anything is purged. Configured per workspace by Sheptra today; a customer-facing setting is not yet available.

SaaS production defaults: thirty days of orchestration data, 180 for Optimize, seven for Zeebe, negotiable on Enterprise. Self-Managed sets it in the Helm chart, thirty days by default.

No configurable retention window found in the docs. A Case Deletion API (v2026.3) removes a case and its Requests on demand, logged to the security log.

Sign-in through WorkOS AuthKit; SSO and directory sync configured from the hosted admin portal linked off the Members page, with memberships mirrored from directory webhooks.

SaaS: SAML or Entra ID on the Enterprise plan, set up through a support ticket. Self-Managed: OIDC. SCIM is claimed on the platform page; no documentation page for it was found.

SAML plus Auth0, Microsoft, Google, Keycloak, GitHub and Atlassian logins; SCIM provisioning of users and groups from Okta or Entra ID. Professional and Enterprise plans.

Two trails: every operator command against the engine, and every administrative action on the organization, each filterable by actor, action, target and time, with before / after diffs and actor names denormalised so rows survive a member leaving.

The 8.9 audit log records process, identity and user-task operations — who, when, what — on SaaS and Self-Managed; Console adds an exportable organization activity log.

Security Logs record ninety-plus event types — logins, SSO and SCIM changes, process publishes, task reassign, retry and rollback, API tokens — downloadable. Enterprise plan.

Managed cloud only, operated by Sheptra. Every workspace runs on a shared cell today; a dedicated database, dedicated compute and a deployment into your own cloud account, still operated by Sheptra, are on the roadmap. No self-managed edition is offered or planned.

SaaS on dedicated single-tenant clusters (GCP or AWS, PrivateLink for Enterprise); Self-Managed with Helm on Kubernetes, Docker, or plain JARs.

Primarily an enterprise cloud offering on AWS or Azure. The open-source edition installs on-premises (PHP, MySQL, Docker) without commercial support; no Kubernetes guide, and enterprise on-premises is not documented.

Not yet available for purchase. Four planned tiers — Developer (free sandbox), Team, Business and Enterprise — priced on environments, seats and history rather than instance volume. Prices will be published once set.

Since 8.6 the core components are source-available under the Camunda License 1.0: free for non-production use, an Enterprise licence for production. Web Modeler free up to five users; the connector SDK is Apache 2.0.

Professional and Enterprise plans gate features throughout the docs; pricing is per case (one per process start, unlimited users), by quote. The public pricing page now redirects to a higher-education package.

Encryption at rest and in transit, SSO and audit trails today, with a published security overview and sub-processor list. SOC 2 Type II, an independent penetration test and a counsel-reviewed DPA are planned.

Trust Center lists SOC 2 Type II, ISO 27001, GDPR, TISAX and the Cloud Security Alliance. No penetration-test or HIPAA statement on the public page.

SOC 2 Type II reports under NDA, ISO 27001, an ISO/IEC 42001 framework, a GDPR DPA and annual penetration tests. HIPAA is not stated.

Drive dashboards: running, incident and completed counts with per-process rollups, plus a statistics endpoint for usage. No process-mining or reporting product on the near-term roadmap.

Optimize: reports, dashboards, heatmaps and alerts over exported process data. A separate component; production use needs a licence.

Embedded Process Analytics — overview, analysis and trend reports, KPIs, CSV export — plus dashboards. Professional and Enterprise plans.

Not offered. Desktop automation would run in a worker you operate.

Camunda RPA: Robot Framework scripts run by a standalone RPA worker, documented for 8.7 and later on SaaS and Self-Managed. The page does not state a maturity level.

An integration with UiPath — robots and jobs monitored from ProcessMaker — rather than a native bot runtime. Document extraction comes from the IDP connector on Enterprise.