Everything below runs today; items marked Pilot support are delivered with a pilot. The desktop app and the server deployment share the same service layer, so a project started on a laptop behaves identically on the ops network.
Steriod UI (browser) Steriod App (desktop)
│ REST │ native bindings
▼ ▼
Service Manager — projects · suites · runs · pipelines
│ gRPC │
▼ ▼
Axiom engine (compiler + runtime) Trend store (TimescaleDB)
telemetry in:
spacecraft / simulator → CCSDS/UDP → YAMCS → MQTT → pipelinesA domain-specific language for temporal, behavioural rules over data streams. Human-readable enough for an auditor, formal enough for a compiler.
The Axiom compiler and reference runtime are written in Go and exposed as a gRPC service, a CLI, and a WebAssembly build — the same engine you try in the Playground.
libaxiom: strict C99, zero malloc, no recursion, bounded loops, no syscalls. A differential test proves its events identical to the Go and Rust runtimes. See the Embedded Rule Runtime.
A language server (LSP) and a VS Code extension: syntax highlighting, diagnostics, and completion for .axiom files.
The engine records evaluations and signal data as it runs — into files or TimescaleDB — so every verdict is reconstructible after the fact.
Source, transformer, rule evaluator, notifier and statistical analysis, plus jobs that mine envelopes, drift, joint models and command-to-response maps on the mission clock. Every mined result is staged for human review before it can alarm.
Document import, rule drafting, root-cause analysis and signal analysis are declared workflows. Each step says whether a model runs in it; engine steps never call one. Every run is fully audited and can be pinned to a different model to cross-test.
Signals, types and enumerations imported from XTCE, so every rule is type-checked against the catalogue before it runs.
All business logic lives in the service backend. The desktop app and the browser UI are two faces of the same system — no feature exists in one that can't be reached from the other.
The backend: projects, test suites, requirements, runs, dashboards, pipelines, settings — with pluggable storage and a REST API. This is where the approval gate lives.
A cross-platform desktop app embedding the same engine and an MQTT broker — a self-contained V&V workstation for a single engineer or a test bench.
The web frontend: coverage and approvals, rule editing, the live rule board and episodes, dashboards, trend charts, the system map and the Vi assistant.
| Component | What it does |
|---|---|
| YAMCS integration | A YAMCS plugin re-exports telemetry to MQTT, feeding Steriod pipelines from the mission-control system many smallsat teams already use. |
| Trend store | TimescaleDB-backed storage for recorded runs and live telemetry — the substrate for trend analysis, multi-run comparison, and RCA timelines. |
| Signal metadata | XTCE and ARXML readers give rules and analyses real signal names, units, and calibrations instead of raw packet offsets. |
| SCOS-2000 MIB importPilot support | The mission information base of SCOS-2000, CCS5 and EGS-CC centres imported as the signal vocabulary. |
| Recorded runs from filesPilot support | CSV and TDMS post-test files and CCS5 retrieval exports replayed through the same pipelines as the live feed. |
| Satellite simulator | A CCSDS packet simulator with scriptable anomaly injection — rehearse fault scenarios and produce labelled test data end to end. |
| Transports | MQTT and Kafka-compatible streaming (Redpanda) in the stack today, plus JSONL telemetry archives; the pipeline layer is transport-pluggable. |
| Deployment | The full platform ships as a Docker Compose stack (UI, backend, engine, trend store, brokers, optional YAMCS) on a single on-prem node, with an offline bundle for air-gapped sites. |
The same tool layer is exposed four ways, so both your engineers and your AI tools work against identical capabilities.
Connect Claude or any MCP-capable agent to your Steriod instance: browse projects, query signals, validate rules, run analyses. Deliberately, there is no approve tool — sign-off stays human.
The identical tool set as a command-line client for scripting and CI, and as an HTTP service — including document conversion (PDF, XTCE, office formats → structured markdown).
Curious how the AI layer uses this? See the agents page →
We can stand up the full platform against a recorded telemetry set in a single session.