A working Rust backend, not yet a qualified production release. Initial scope is single-node SQLite; Aira and deadline capabilities are not implemented.
Contact by emailBinean Echelon
Echelon hosts durable orchestration across services and human work, with AI participation planned. A shared model describes the work and its result; identity, authorization and execution protocols remain distinct. This page separates implemented capabilities from design goals.
The problem
A case may cross service calls, human review and long waits. When each tool keeps only its own fragment, it becomes difficult to see what is pending and who can act.
An orchestration backend makes that journey explicit. Echelon targets processes that need durable state and accountable decisions—not every short API call needs an engine.
Architecture
Each layer has a defined responsibility; none should silently take over another’s decisions.
The model
Task and Outcome concepts allow a common description of work. They do not make a service credential interchangeable with a person, nor establish a universal wire protocol. A valid JSON result can still be factually or operationally wrong.
Responsibility
People, services and AI have different strengths. Human judgment and accountability are not inefficiencies to remove.
Operational history is evidence of execution, not automatically clean training data or permission to train. Any reuse needs consent or another appropriate basis, access controls, curation and quality evaluation.
Changing who performs work requires validation of the result and failure handling. Flow versioning makes a change identifiable; it does not make autonomous optimization safe or authorize it. Self-learning and automatic process optimization are not shipping capabilities.
Design choices
SPINE defines behavior; Echelon implements and hosts it. This helps keep UI behavior and execution responsibilities aligned.
Tradeoff: specifications, implementation and integration evidence must stay aligned. A written contract is not release qualification.
Human-task surfaces and external worker integration are implemented. Vista supplies human-facing forms and administration.
Tradeoff: identity, authorization and integration testing remain necessary for each execution path.
The current Rust runtime packages durable execution with SQLite. Storage abstraction is an architectural boundary, not a menu of supported database backends.
Tradeoff: initial scope is single-node. Arbitrary Postgres or Kafka deployment, HA and multi-tenancy are not current claims.
Versioned Flow publication makes the selected definition explicit. A new version is a change to review, not an invisible update.
Tradeoff: versioning alone does not validate business outcomes, migrate live state safely or undo external effects.
Applications
Candidate uses include case approvals, operations and AI-assisted review. These are use cases, not customer deployments. BENOVA is intended to apply EVA to Ingenium alongside Nexus and Orbit.
Once stable, EVA is intended to orchestrate work on VISTA itself: request, human scope approval, AI proposal or patch, machine validation, independent review and human acceptance. Review must identify the exact artifact and version.
This self-hosting direction is not live functionality or a promise of automatic self-modification, merging or deployment. Software work is a proving ground; Echelon remains a general-purpose orchestration backend.
Alternatives
These products address overlapping but different needs. The links describe their own capabilities; they are not endorsements of Echelon. Echelon is not yet equivalent in production maturity.
| Product | Focus and selection considerations | Official documentation |
|---|---|---|
| Binean Echelon | Rust backend with SQLite, versioned Flow publication, inspection, human-task surfaces and external workers. Single-node initial scope; qualification remains incomplete. Evaluate as a bounded pilot, not a mature substitute. | |
| Temporal | Durable execution using Workflows and Activities. Evaluate its execution model, worker operations and versioning approach against your code and deployment lifecycle. | |
| Camunda 8 | BPMN-oriented orchestration with Tasklist for human work. Consider it when process modeling and human-task operations are central, and assess integration and deployment requirements. | |
| AWS Step Functions | Managed state-machine orchestration in AWS, including immutable state-machine versions. Assess AWS integration, execution semantics and pricing for the specific workload rather than assuming negligible cost. | |
| LangGraph | Graph-based orchestration for stateful agents, with durable execution and human-in-the-loop capabilities. Assess it for AI-centric workflows; do not assume human approval or persistence is absent. |
Selection depends on the process model, human work, integration ecosystem, operating constraints and support needs. Test representative failures and changes, not just a happy-path demonstration.
For a near-term production requirement, start by evaluating qualified, established options. Echelon offers a model to investigate, not proven uniqueness or a production-maturity shortcut.
Conditions of use
Status
Implemented: a Rust backend with durable SQLite runtime, HTTP interfaces, versioned Flow publication, process and task inspection, external worker integration and human-task surfaces. Release qualification is not complete.
Vista has working Flow inspection and editing, Human Task forms and acquisition, user administration and process activity. Its Process workspace is still being completed, with integrated qualification ongoing.
Aira remains in design with no Rust implementation or settled public protocol. Scheduler and Timeout are not implemented and their contracts remain open. Private deployment is a target profile, not a compliance certification.
Bring a bounded process, its failure cases and the evidence your team needs before operational use. We can discuss fit without treating a prototype as a production commitment.
