Orion

Build services at AI speed on a consistent, governed foundation.

Define the business logic in JSON. Orion supplies the runtime around it.

Orion is a declarative services runtime. An Orion service consists of a workflow and a channel, with optional connectors to external systems. Post those definitions, activate them, and the service is live—without writing an application server, building a container for each service, or restarting the runtime.

Orion consistently handles the surrounding infrastructure: routing, ingress guards, rate limits, timeouts, retries, circuit breakers, connection pooling, version rollout, rollback, metrics, and tracing. Definitions can be written by a developer or proposed by an AI assistant; they follow the same draft, test, activation, and audit path.

[!NOTE] Orion is designed for request- and event-shaped work expressible as ordered task functions and JSONLogic. It is not a general-purpose application runtime or a durable workflow engine. See When Orion fits before choosing it for a project.

Quickstart

This tested path uses Docker, curl, and a POSIX-compatible shell. It starts a local Orion instance, deploys an order-processing workflow and channel, and calls the resulting API.

1. Start Orion

docker run --name orion-quickstart -d -p 8080:8080 \
  ghcr.io/goplasmatic/orion:latest

curl --retry 10 --retry-delay 1 --retry-connrefused \
  http://localhost:8080/healthz

2. Inspect and deploy the example

curl -fsSLo /tmp/orion-quickstart.sh \
  https://raw.githubusercontent.com/GoPlasmatic/Orion/main/examples/quickstart.sh
less /tmp/orion-quickstart.sh
bash /tmp/orion-quickstart.sh

The script makes four administration calls: it creates and activates a workflow, then creates and activates its channel. It finishes with a test request and is safe to run again. From a cloned repository, run ./examples/quickstart.sh instead.

3. Call your service

curl -fsS -X POST http://localhost:8080/api/v1/data/orders \
  -H 'Content-Type: application/json' \
  -d '{ "data": { "order_id": "ORD-0001", "total": 12500 } }'

The response contains the parsed order with "flagged": true. The workflow holds the business logic; the channel exposes it at POST /orders. Routing, validation, tracing, and lifecycle management are supplied by Orion.

For expected output, troubleshooting, Windows-friendly installation paths, and cleanup instructions, follow the complete quickstart.

The core model

flowchart LR
    Request["HTTP request, Kafka record, or schedule"] --> Channel
    Channel["Channel<br/>route + ingress policy"] --> Workflow
    Workflow["Workflow<br/>ordered business logic"] --> Connector
    Connector["Connector<br/>external system"]
    Workflow --> Response["Response or stored trace"]

The definitions belonging to one service can be shipped together as a package. One Orion instance can run many packages side by side, each with its own lifecycle and rollout.

How Orion works explains request execution, composition, deployment topology, and extension boundaries in one page.

What you can build

• Microservice and decision APIs: expose transformations, validation, pricing, eligibility, and routing decisions over HTTP.
• Webhook and data-ingestion services: normalize incoming payloads and read or write external systems through governed connectors.
• Kafka consumers: process records, publish results, and route failed messages to a dead-letter topic.
• AI-agent tools: expose governed HTTP operations that an assistant can draft, dry-run, activate, and roll back using the Orion CLI and agent skill.
• Scheduled jobs: run a workflow on a cron expression instead of on a request, with every scheduled instant recorded as a durable occurrence.
• Composable services: call another channel in-process with channel_call, while preserving the callee's guards and preventing cycles.

Choose a focused path in What are you building?, or deploy a tested package from examples/packages/.

Runtime capabilities

Configuration is explicit, and several production controls are permissive or disabled for local development. In particular, data channels are open unless they declare authentication or sit behind an authenticating proxy. Before exposing an instance, work through the production checklist.

Safe changes, including AI-generated ones

flowchart LR
    Author --> Draft --> Validate --> DryRun["Dry-run"] --> Activate --> Rollout --> Observe
    Observe -->|problem| Rollback

Workflow, channel, and connector definitions use the same governed lifecycle regardless of who authored them. A draft serves no traffic. You can validate and dry-run it, approve the exact version, activate it without restarting the server, roll traffic out by percentage, and return to a previous immutable version. Administrative changes are recorded in the audit log.

For AI-assisted authoring, install the Orion agent skill or use the self-contained prompt pack. The Claude Code tutorial walks through a complete assisted workflow.

When Orion fits

Orion fits services whose work starts with an HTTP request or Kafka record, completes as a bounded pipeline, and can be expressed with Orion's task functions and JSONLogic. It is especially useful when many small services need the same operational and governance foundation.

Choose another runtime, or pair one with Orion, when:

• work must survive restarts at an intermediate step or wait hours or days;
• ingress requires gRPC, WebSockets, or streaming responses;
• business logic requires arbitrary code with I/O, or a scripting runtime — a pure transformation ships as a sandboxed WebAssembly plugin, anything that has to reach another system does not;
• image processing, model inference, or large in-memory joins sit on the hot path; or
• full OIDC flows or mutual TLS must terminate inside the data plane. JWT verification is built in; those flows require a gateway or service mesh.

Read Is Orion right for you? for detailed comparisons with durable execution engines, API gateways, automation platforms, rule engines, and embedded dataflow-rs.

Install

The current workspace release is 1.7.0 and requires Rust 1.98 when built from source. The server and CLI are released in lockstep; use matching versions.

# Homebrew: macOS Apple Silicon and Linux
brew install GoPlasmatic/tap/orion-server
brew install GoPlasmatic/tap/orion-cli

# Server from source
cargo install --git https://github.com/GoPlasmatic/Orion --locked orion-server

# Server and CLI from source
cargo install --git https://github.com/GoPlasmatic/Orion --locked \
  orion-server orion-cli

Release installers are also available for Linux, macOS Apple Silicon, and Windows. See Install & Run for every method, platform support, startup, and verification.

Useful local commands:

orion-server validate-config -c config.toml
orion-server fmt ./definitions
orion-server lint workflow.json
orion-server clippy ./definitions
orion-server dry-run -w workflow.json -i input.json
orion-server test examples/workflow-tests
orion-server compile ./definitions -o package.json
orion-server package apply -s http://localhost:8080 -f package.json

The server command reference and Orion CLI reference list all commands and flags.

Performance

The published Orion 1.0.0 benchmark measured 5.1K–5.7K workflow requests per second on one Apple M2 Pro instance, with single-digit-millisecond average latency across the recorded workflow scenarios. These are release- and workload-specific results, not capacity guarantees. Review the benchmark record for hardware, scenarios, tail latency, cluster results, and reproduction steps.

Documentation

The full manual is at docs.goplasmatic.io.

The documentation source lives in docs/src/. Edit that directory, not the generated docs/book/ output. See docs/STYLE_GUIDE.md for its structure and editorial conventions.

Project and community

• Orion UI provides the browser-based operations console.
• orion-cli manages workflows, channels, connectors, plugins and schedules from a terminal.
• CONTRIBUTING.md covers development setup, tests, and pull requests.
• Use GitHub Discussions for questions and GitHub Issues for bugs.
• Report vulnerabilities privately according to SECURITY.md.
• If you use Orion, add your project to ADOPTERS.md.

Orion is built by Plasmatic and released under the Apache License 2.0.