Palari Brain 0.1.0-alpha.1

Palari Brain Alpha

Memory that can show its work.

This is the first public developer preview of Palari Brain, an evidence-first
memory kernel for chat assistants.

Install

npm install github:CoyStan/palari-brain#v0.1.0-alpha.1

You can also download and install the attached
palari-brain-0.1.0-alpha.1.tgz. Verify it with the attached SHA256SUMS file.

Highlights

• Canonical dialogue with speaker, time, scope, and stable identity.
• Bounded recall that resolves back to exact source evidence.
• Evidence-bound answers with explicit uncertainty.
• Correction, exact deletion, and user/workspace isolation.
• Provider-neutral core with replaceable model, embedding, and reranker seams.
• Bounded native Ettin reranking and optional shared local-worker pacing.

Validation

• Core suite: 106 passed, 0 failed.
• Full legacy suite: 995 passed, 3 optional skips, 0 failed across 998.
• Quickstart: 6 of 6.
• Production dependency audit: 0 vulnerabilities.
• Real tarball clean-install and public API import: passed.

The latest-attempt S60 alpha diagnostic produced 56 correct and 4 incorrect
answers across 60. A harder metadata-selected diagnostic produced 17 judged
correct, 2 judge failures, and 1 execution failure; manual directional review
found 19 correct and 1 incorrect. These are alpha diagnostics, not one-shot
release benchmarks.

Known limitations

• One observed candidate-review state error can stop an answer before delivery.
• Later retrieval can sometimes treat future intent as completed action.
• File-backed worker pacing is local to one host.
• Node.js 22 can emit an experimental SQLite warning.
• This package is not published to the npm registry.

See CHANGELOG.md
for the full release notes.