Knolo Cortex: Verifiable Traces Over Knowledge Images

Knolo Cortex is the verifiable-trace side of the Knolo stack: bounded evidence retrieval over a mounted V4 pack or V5 Knowledge Image, with roots you can recompute later. It ships in @knolo/core 6.2.0 as the separate import @knolo/core/cortex. This note explains what a Cortex trace is for, how verification is scoped, and how that fits the open Knolo contract on Knolo on knolo.org—without treating a passing check as a truth verdict.

What Knolo Cortex adds to Knowledge Images

Knolo is an open-source, local-first Knowledge Image runtime and format. You pack documents into a portable .knolo artifact, pin it by digest, and mount it with @knolo/core. Ordinary lexical query remains the everyday path. Knolo Cortex sits beside that path as reasoned evidence retrieval: normalize a question, run retrieval rounds against the mounted image, optionally let a host-supplied model select among bounded excerpts, then emit an answer tied only to selected evidence—and a trace that records the loop.

Import matters. Knolo Cortex is createCortexSession from @knolo/core/cortex. The root @knolo/core entry does not load it. A differently named helper, createCortex on the root export, is the memory overlay (remember / recall); do not mix the two APIs. The Knolo Cortex docs state the split explicitly.

Knolo Cortex stays local-first: the host owns the model

Core does not call a network, does not embed a model, and does not decide that an answer is true. The host owns any model. The model sees the question, hit ids, and short excerpts (docs cap excerpts at 500 characters)—not the whole corpus. That keeps Knowledge Images portable: evidence stays in the mounted bytes you control, and the trace is something you can store next to a receipt instead of a chat log you cannot replay.

Rust, Python, and ICP runtimes do not implement Cortex today; they stay on the 5.5.1 line while TypeScript @knolo/core 6.2.0 carries the Cortex session API. Cite registry versions as published; do not assume every language surface ships the same features on the same day.

What Knolo Cortex verification means

verifyCortexAnswer recomputes traceRoot, answerRoot, and evidenceRoot, checks evidence identity, and replays the retrieval receipt against the image. Evidence must come from a replayed round. A dossier, when present, must match the trace body. A passing verification means the trace matches the declared evidence. It does not mean the claim is proven, the source document is correct, or the model answer is right.

That boundary matches the broader Knolo verification rule on knolo.dev verification: structural integrity is not a truth verdict. Roots and receipts describe how domain-separated digests bind encoded state so later inspection can recompute the same roots. Cortex extends that receipt story to a multi-round evidence loop.

How Cortex fits Knolo open technology

Cortex is a runtime feature; the public record around the format and naming lives on knolo.org:

• Knolo specifications describe the Knowledge Image / .knolo contract (V5 current in source, V4 kept for compatibility). Numbered adopted specification records are not all published yet; unpublished stays unpublished on purpose.
• Knolo licensing orients readers to Apache License 2.0 for Knolo Core, with trademarks kept separate from the software license.
• Knolo compatibility treats independent implementation as allowed under the open terms—compatibility is not endorsement, and “Knolo Certified” is a separate program.
• Knolo Hub remains a pin-and-mount registry. Hub does not run Cortex sessions for you; you mount the image locally, then run the session in a host you control.

Together those pages keep the open-technology promise: portable artifacts, inspectable roots, and clear boundaries between software rights, format contracts, and identity.

Practical next steps with Knolo Cortex

1. Read the product overview on Knolo on knolo.dev and the Cortex docs.
2. Install @knolo/core 6.2.0, mount a V4 pack or V5 image, and create a session with createCortexSession from @knolo/core/cortex.
3. Run askCortex, store the trace, and check it with verifyCortexAnswer—treat a pass as “roots match,” not “claim proven.”
4. For specifications, licensing, and verification domains, use Knolo on knolo.org.
5. Source and releases: the Knolo core repository on GitHub.

Knolo Cortex exists so evidence retrieval can leave a replayable trail: Apache-2.0, local-first, host-owned models, and verification that stays honest about what it proves. Pair the runtime docs with the open-technology pages above when you describe what Knolo ships today.