Back to projects
nage / Knowledge and model operations

Making the system behind an answer visible.

nage is a platform project connecting knowledge ingestion and versioning, access boundaries, model execution and records of how a result was produced. This portal brings architecture, implementation and public resources into one guide.

Work statusControlled beta and developmentSources checked
Official website
01

Why nage?

A model response is one part of a larger AI workflow. Sources enter the system, are selected within a version and access scope, participate in an operation, and inform a result passed to another application. nage explores how knowledge and the record of that operation can be managed along this path.

Knowledge with its source.
Origin, version and permitted scope are considered separately from the answer text.
A record for the next recipient.
The application, operator and reviewer may need different disclosure views of the same record.
Responsibilities before names.
Model architecture, operating environment and record contract have distinct jobs; one does not stand in for another.
02

System responsibilities

Four distinct responsibilities

Operating environment

nage AI OS

Addresses sources, state, access, transport and observation across operating components.

Reference architecture

SEDIM

Defines architectural roles for source-aware results and their formation records. The trained-layer path is evaluated separately from current grounding.

Candidate record standard

KLM

A contract project describing source, signal and assessment disclosure, rather than a certification scheme.

Research and development

Craton / Pro

Investigates whether managed layers can provide usable knowledge. It is not another name for the current service path.

From knowledge to review

  1. 01

    Source and scope

    Identify the document version, origin and access boundaries governing its use.

  2. 02

    Execution and context

    Record which sources and methods the selected execution path declares.

  3. 03

    Answer and record

    Carry the result, disclosed signals and reasons for missing information to the next recipient.

This diagram describes responsibilities, not a verified live service-call graph. Each boundary must be tested in the selected deployment.

Practical questions

What if a policy document changes?
Identify the version behind the earlier answer, the records needing review and the checks to rerun.
How can a claim be reviewed?
Read the source reference, the evidence offered for the claim and the method that produced that evidence together.
What if another application receives the record?
Preserve a missing measurement’s null and reason, and specify which fields the recipient must recheck.
03

From design to use

A small, bounded record review can be a better starting point than a large deployment promise. Public developer materials provide a synthetic fixture and verification package for this purpose.

  1. 01

    Inspect the same synthetic fixture in Record Lab.

  2. 02

    Download the verification package and read its README requirements.

  3. 03

    Check the sample locally and observe what happens when it changes.

  4. 04

    For actual use, separately define data scope, key trust and acceptance criteria.

Check the synthetic sample locally

node verify.ts sample/receipt.json --pubkey sample/pubkey.txt

Extract the verification ZIP and run this in the verification folder with Node.js 24. Its sample and test key do not authenticate a production service or a real-world signer. This page neither runs the command nor sends data to the service.

04

Related writing

05

Documents and links

06

Status and limits

The public product documents reviewed on 22 October 2026 describe the current service path as Qwen3-8B grounding and record generation. Trained VARVE/CLAST layers and Craton inference remain development and research work. Controlled beta does not mean the entire target architecture is complete.

Record Lab and the downloadable verification package use a synthetic fixture and a public test key. Production-chain boundaries—including HMAC, an unsigned Merkle root and the open CONDUIT Ed25519 integration—are disclosed separately. Removing a source from serving does not erase every stored artifact or knowledge already present in a base model.

Record integrity, signature validity, claim support and evidence that execution occurred are different questions. Success on one does not establish the others.

This portal is a reading map of public documents, not a service acceptance test, performance measurement or independent audit result.

Existing project text and dated sources

The project record edited in the administration panel. Its date-pinned links identify the revision used for that text; the directory above uses current repository paths.

The system behind an answer

What must an AI service manage beyond producing a model response? nage is a platform project connecting ingestion, versioned knowledge, authorised context selection, model execution and the trace of how a result was produced.

Distinct responsibilities

nage denotes the operating platform and API service. KLM defines knowledge, provenance, evaluation and verification contracts; SEDIM denotes the model architecture and CRATON the model family. Naming one component does not establish acceptance of another.

Related open contract work: KLM.

Design and implemented scope

The main document dated 22 September describes a target system, not a completed implementation. Compat, Native Core and Native Pro are target service profiles, not commercial pricing tiers.

The platform README reviewed on 14 October states that the general backend bootstrap remains disabled until a validated schema can be reproduced. A narrow upload acceptance environment is separate. Success there does not establish live acceptance of the whole platform.

Next evaluation

The source lifecycle, access isolation, replay, quality and total cost need evaluation together, on the same version and workload. This entry does not claim general availability, completed product acceptance or a performance advantage.

Source note

Sources reviewed on 14 October 2026: Nage AI OS main source v1.3 (22 September) and the nage-platform README. No service, customer data or new acceptance test was run for this page.

07

Atlas and KLM

KLM

Continue to the KLM portal for provenance, record contracts and validation tools.

The public KLM charter on the NAGE site and the open-source KLM 0.3 tools described here are distinct documentation lines. Read each against its own sources rather than treating their versions as equivalent.