Decide whether to engage
Seven browser-local questions create a private planning record.
Fit Check →Resources
Decision aids, not a content treadmill.
Use the shortest object that answers the question: a local tool, a checklist, a public registry, a method or a dated evidence note.
Start with the task
Compare a supplier
A printable control, evidence and exit checklist.
Vendor checklist →Plan a system
Reference envelopes, model and hardware registries, and planning calculators.
Reference architectures →Review trust
Current controls, responsibilities, non-claims and procurement path.
Trust Centre →Evaluate a model
A reproducible method for performance, quality, refusal and leakage.
Performance method →Govern use
AI literacy, human authority, source handling and operating roles.
Governance guide →Dated evidence notes
Deployment boundaries
Why residency, tenancy, administration, keys, support, backup and exit are separate decisions.
Read note →Model memory planning
Why weights, KV cache, context, batching, runtime and headroom all matter.
Read note →Cited answers
What a source-grounded answer must preserve before a professional can inspect it.
Read note →Governed agents
How identity, tools, budgets, gates and receipts limit delegated action.
Read note →Kimi K3 evaluation
Why available weights unblock evaluation but not production claims.
Read note →AI heat recovery
The measurements needed before waste heat becomes a credible project.
Read note →Model and machine-readable resources
Open-model programme
Compare current open-source and open-weight candidate families by use case, licence class, deployment scale and proof status.
Explore candidate families →Machine-readable site guide
A concise map of the public site for retrieval systems. It makes no crawler, ranking or citation promise.
Open llms.txt →Each note states a reviewed date and evidence boundary. ANULUM publishes when an object adds a durable decision aid, not to manufacture activity.