Status, roadmap and contributing
Where Lumis SDK is, what comes next, how to migrate from earlier versions, how to get in touch, and the research it comes from.
Status
Lumis SDK 0.1.0 is the first release of the current architecture. It covers the investigation part of a larger research direction: read-only incident context, deterministic triage, one optional investigator, mechanical assessment and audit records. It is experimental: useful for investigation and research today, not a production-hardened or generally validated tool.
| Done | Next |
|---|---|
| YAML-led graph and discovery; Kubernetes, Prometheus, Loki, Tempo, Prefect, SQL and change sources | Evaluation with other model families |
| Deterministic triage with strict sufficiency rules | Smaller cookbooks: a single web service, a single data pipeline |
| One bounded investigator with four provider adapters | recent_changes_affecting semantics and a connector conformance kit |
| Mechanical assessment, competing-root abstention, audit store | An independent security, performance and usability review |
| Live evaluation on GridCast with one model | Human verification of the evaluation's mechanism labels |
Not planned for the SDK at this stage: automatic remediation, automatic rule learning, or a hosted service. See the roadmap ↗ and changelog ↗.
Get in touch
Lumis is early and small. Questions, ideas, bug reports and offers to help are welcome by email: solomon@qadimlabs.com ↗.
What is open source
This SDK is open source under Apache-2.0, and its source is on GitHub ↗. It is a proof of concept of Lumis' investigation core, released so that the approach can be inspected, run and evaluated. Other proofs of concept may be published the same way in future.
That does not make all of Lumis open source. Other Lumis products and services, including anything hosted or commercial, are separate and may be offered under different terms.
Migrating from 0.0.x
0.1.0 replaces the earlier framework entirely. The old diagnose, resolve, rules, plugins, memory and lifecycle interfaces were removed, with no compatibility layer. To migrate, start a fresh project with lumis init, map your incidents, identities, topology and observations to the current contracts, and re-express diagnostic rules as checks with predictions and falsifiers. Use a fresh audit store.
The previous implementation is preserved on the legacy branch ↗.
Source: Migration guide ↗ in the SDK repository.
Research
Lumis grew out of Agentic Self-Healing for Data & AI Pipelines: An Affordable Vendor-Agnostic Architecture using Open-Source Software ↗ (arXiv 2608.01955, preprint), also available as a PDF. The paper describes a broader architecture that includes guarded recovery and verified learning. The SDK implements the investigation part only.