hippo-memory 1.70.0 → 1.72.0

package/dist/importers/markdown-parse.d.ts

@@ -1,32 +1,18 @@
-/** Minimal inline frontmatter split. Recognises a leading `---\n…\n---\n`
- *  block (no YAML dep). Returns the parsed key→value map plus the body with the
- *  block removed. When no well-formed block is present, `fm` is empty and
- *  `body` is the original content. */
+/** Minimal inline frontmatter split (no YAML dep) on a leading `---\n...\n---\n` block; returns the key-value map plus the body,
+ *  or an empty `fm` and the original content when no well-formed block is present. */
 interface FrontmatterParseResult {
     fm: Record<string, string>;
     body: string;
 }
 export declare function splitMarkdownFrontmatter(raw: string): FrontmatterParseResult;
-/** Pull a frontmatter field that may be a YAML flow list (`[a, b]`), a
- *  comma-separated scalar (`a, b`), or a single token, into a string[]. Quotes
- *  and surrounding brackets are stripped; empty entries dropped. */
+/** Pull a frontmatter field that may be a YAML flow list (`[a, b]`), a comma-separated scalar, or one token into a string[];
+ *  quotes and brackets are stripped and empty entries dropped. */
 export declare function frontmatterList(value: string | undefined): string[];
-/** Parse `[[wikilinks]]` from body text. `[[target]]` and `[[target|alias]]`
- *  both yield `target` (alias dropped). Returns de-duplicated, order-preserving
- *  target strings (trimmed). Embeds (`![[…]]`) are intentionally matched too —
- *  the leading `!` is not part of the `[[…]]` capture, so an embed contributes
- *  its target as a candidate, which is the desired no-crash baseline behaviour. */
+/** Parse `[[wikilinks]]` from body text: `[[target]]` and `[[target|alias]]` both yield `target`; de-duplicated, order-preserving.
+ *  Embeds (`![[...]]`) match too, because the leading `!` is outside the capture and a candidate target is harmless. */
 export declare function parseWikilinks(body: string): string[];
-/** Recursively collect files whose name matches `match` (`*.md` by default) under `root`,
- *  as paths relative to `root` with forward-slash separators (stable artifactRef keys across OSes).
- *  Symlinks are not followed. Skips dot-directories (the default `.hippo` store,
- *  `.git`, `.obsidian`, `.trash`) AND the canonicalized Hippo store path during
- *  the walk, so re-importing a vault that CONTAINS the store never ingests its own
- *  markdown mirror files (codex R5 P1: `hippo import --vault .` after `hippo
- *  init` in the vault would otherwise self-import its mirror rows and grow on
- *  every run). The root-IS-the-store case is handled one level up in
- *  importVault (a no-op early return), NOT here: returning [] for it would feed
- *  the deletion-sync an empty scan that mass-archives every live row (codex R8). */
+/** Recursively collect files matching `match` (`*.md` default) under `root` as forward-slash relative paths; symlinks are not followed.
+ *  Skips dot-directories and the Hippo store path so a vault containing the store never imports its own mirror files; importVault handles root-IS-the-store. */
 export declare function collectMarkdownFiles(root: string, hippoRoot: string, match?: RegExp): string[];
 export {};
 //# sourceMappingURL=markdown-parse.d.ts.map

package/dist/importers/markdown-parse.js

@@ -16,9 +16,7 @@ export function splitMarkdownFrontmatter(raw) {
             continue;
         let val = kv[2].trim();
         if (val === '') {
-            // YAML block-style list: `key:` followed by indented `- item` lines
-            // (common in Obsidian/Dendron frontmatter). Collect them into a
-            // comma-joined value so frontmatterList parses them (codex P2).
+            // YAML block-style list (`key:` then indented `- item` lines, common in Obsidian/Dendron): collect into a comma-joined value for frontmatterList.
             const items = [];
             let j = i + 1;
             let item;
@@ -35,9 +33,8 @@ export function splitMarkdownFrontmatter(raw) {
     }
     return { fm, body };
 }
-/** Pull a frontmatter field that may be a YAML flow list (`[a, b]`), a
- *  comma-separated scalar (`a, b`), or a single token, into a string[]. Quotes
- *  and surrounding brackets are stripped; empty entries dropped. */
+/** Pull a frontmatter field that may be a YAML flow list (`[a, b]`), a comma-separated scalar, or one token into a string[];
+ *  quotes and brackets are stripped and empty entries dropped. */
 export function frontmatterList(value) {
     if (!value)
         return [];
@@ -49,11 +46,8 @@ export function frontmatterList(value) {
         .map((s) => s.trim().replace(/^['"]|['"]$/g, '').trim())
         .filter(Boolean);
 }
-/** Parse `[[wikilinks]]` from body text. `[[target]]` and `[[target|alias]]`
- *  both yield `target` (alias dropped). Returns de-duplicated, order-preserving
- *  target strings (trimmed). Embeds (`![[…]]`) are intentionally matched too —
- *  the leading `!` is not part of the `[[…]]` capture, so an embed contributes
- *  its target as a candidate, which is the desired no-crash baseline behaviour. */
+/** Parse `[[wikilinks]]` from body text: `[[target]]` and `[[target|alias]]` both yield `target`; de-duplicated, order-preserving.
+ *  Embeds (`![[...]]`) match too, because the leading `!` is outside the capture and a candidate target is harmless. */
 export function parseWikilinks(body) {
     const out = [];
     const seen = new Set();
@@ -71,22 +65,11 @@ export function parseWikilinks(body) {
     }
     return out;
 }
-/** Recursively collect files whose name matches `match` (`*.md` by default) under `root`,
- *  as paths relative to `root` with forward-slash separators (stable artifactRef keys across OSes).
- *  Symlinks are not followed. Skips dot-directories (the default `.hippo` store,
- *  `.git`, `.obsidian`, `.trash`) AND the canonicalized Hippo store path during
- *  the walk, so re-importing a vault that CONTAINS the store never ingests its own
- *  markdown mirror files (codex R5 P1: `hippo import --vault .` after `hippo
- *  init` in the vault would otherwise self-import its mirror rows and grow on
- *  every run). The root-IS-the-store case is handled one level up in
- *  importVault (a no-op early return), NOT here: returning [] for it would feed
- *  the deletion-sync an empty scan that mass-archives every live row (codex R8). */
+/** Recursively collect files matching `match` (`*.md` default) under `root` as forward-slash relative paths; symlinks are not followed.
+ *  Skips dot-directories and the Hippo store path so a vault containing the store never imports its own mirror files; importVault handles root-IS-the-store. */
 export function collectMarkdownFiles(root, hippoRoot, match = /\.md$/i) {
     const out = [];
-    // Canonicalize (realpath) so a non-dot HIPPO_HOME store nested in the vault is
-    // skipped even when hippoRoot is an aliased path (junction / Windows case
-    // variant); path.resolve would miss it and self-import the store's mirror
-    // files (codex R9 follow-up: same gap as the importVault guard, sibling site).
+    // Canonicalize (realpath) so a non-dot HIPPO_HOME store nested in the vault is skipped even via a junction or Windows case alias.
     const resolvedHippoRoot = realpathOrResolve(hippoRoot);
     const walk = (dir) => {
         const entries = fs.readdirSync(dir, { withFileTypes: true });

package/dist/importers/markdown.js

@@ -1,13 +1,8 @@
 import * as fs from 'fs';
 import { importEntries } from './core.js';
 const SLUG_MAX_CHARS = 50;
-// ---------------------------------------------------------------------------
-// Structured markdown importer (MEMORY.md / AGENTS.md format)
-// ---------------------------------------------------------------------------
-/**
- * Slugify a heading for use as a tag.
- * "Data Pipeline & Cache" -> "data-pipeline-cache"
- */
+/** Slugify a heading for use as a tag.
+ *  "Data Pipeline & Cache" -> "data-pipeline-cache" */
 function slugify(heading) {
     return heading
         .toLowerCase()
@@ -17,11 +12,7 @@ function slugify(heading) {
         .replace(/-+/g, '-')
         .slice(0, SLUG_MAX_CHARS);
 }
-/**
- * Parse structured markdown into {content, sectionSlug} pairs.
- * Each heading starts a new section. Bullet points / numbered items under
- * the heading become individual memories tagged with the section slug.
- */
+/** Parse structured markdown into {content, sectionSlug} pairs: each heading starts a section whose bullets or numbered items become memories. */
 function parseStructuredMarkdown(raw) {
     const results = [];
     const lines = raw.split('\n');

package/dist/importers/sources.js

@@ -2,6 +2,7 @@ import * as fs from 'fs';
 import * as path from 'path';
 import { importEntries } from './core.js';
 import { splitMarkdownFrontmatter, collectMarkdownFiles } from './markdown-parse.js';
+import { errorMessage, log } from '../util/log.js';
 import { isJsonString, isJsonObject } from '../util/json.js';
 /** Coerce one imported record (string, `{content|text: ...}` object, or
  *  anything else) into the plain-text memory chunk it represents. */
@@ -13,14 +14,7 @@ function extractMemoryText(candidate) {
     }
     return '';
 }
-/**
- * Parse ChatGPT memory export file.
- * Supports:
- *   - JSON array of strings: ["memory 1", "memory 2"]
- *   - JSON array of objects: [{"content": "...", "created": "..."}]
- *   - ChatGPT export format: {"memories": [{"content": "...", "created_at": "..."}]}
- *   - Plain text: one memory per line
- */
+/** Parse a ChatGPT memory export: a JSON array of strings or `{content, created}` objects, `{"memories": [...]}`, or plain text with one memory per line. */
 function parseChatGPTFile(filePath) {
     const raw = fs.readFileSync(filePath, 'utf8').trim();
     // Try JSON first
@@ -36,8 +30,8 @@ function parseChatGPTFile(filePath) {
                 return parsed.map(extractMemoryText).filter(Boolean);
             }
         }
-        catch {
-            // Fall through to plain text
+        catch (err) {
+            log.warn(`import: ${filePath} is not valid JSON (${errorMessage(err)}); reading it as plain text`);
         }
     }
     // Plain text: one memory per line
@@ -47,14 +41,9 @@ export function importChatGPT(filePath, options) {
     const chunks = parseChatGPTFile(filePath);
     return importEntries(chunks, 'import:chatgpt', ['imported', 'chatgpt'], options);
 }
-// ---------------------------------------------------------------------------
-// Claude importer
-// ---------------------------------------------------------------------------
 const HIPPO_START = '<!-- hippo:start -->';
 const HIPPO_END = '<!-- hippo:end -->';
-/**
- * Strip the hippo hook block from markdown content.
- */
+/** Strip the hippo hook block from markdown content. */
 function stripHippoBlock(content) {
     const startIdx = content.indexOf(HIPPO_START);
     const endIdx = content.indexOf(HIPPO_END);
@@ -62,9 +51,7 @@ function stripHippoBlock(content) {
         return content;
     return content.slice(0, startIdx) + content.slice(endIdx + HIPPO_END.length);
 }
-/**
- * Split markdown into meaningful chunks (headings + bullet points).
- */
+/** Split markdown into meaningful chunks (headings + bullet points). */
 function splitMarkdown(content) {
     const chunks = [];
     const lines = content.split('\n');
@@ -119,9 +106,7 @@ function appendLine(chunks, open, line) {
     // Regular line: append to current
     return current ? current + ' ' + trimmed : trimmed;
 }
-/**
- * Parse CLAUDE.md or Claude memory.json.
- */
+/** Parse CLAUDE.md or Claude memory.json. */
 function parseClaudeFile(filePath) {
     const raw = fs.readFileSync(filePath, 'utf8');
     // JSON memory file
@@ -132,8 +117,8 @@ function parseClaudeFile(filePath) {
                 return parsed.map(extractMemoryText).filter(Boolean);
             }
         }
-        catch {
-            // Fall through to markdown
+        catch (err) {
+            log.warn(`import: ${filePath} is not valid JSON (${errorMessage(err)}); reading it as markdown`);
         }
     }
     // Markdown file: strip hippo block and split
@@ -144,13 +129,8 @@ export function importClaude(filePath, options) {
     const chunks = parseClaudeFile(filePath);
     return importEntries(chunks, 'import:claude', ['imported', 'claude'], options);
 }
-// ---------------------------------------------------------------------------
-// Cursor importer
-// ---------------------------------------------------------------------------
-/**
- * Split cursor rules file into chunks.
- * Priority: numbered items, then bullet points, then double newlines.
- */
+/** Split cursor rules file into chunks.
+ *  Priority: numbered items, then bullet points, then double newlines. */
 function parseCursorFile(content) {
     const chunks = [];
     const lines = content.split('\n');
@@ -206,14 +186,7 @@ export function importCursor(sourcePath, options) {
     });
     return importEntries(chunks, 'import:cursor', ['imported', 'cursor'], options);
 }
-// ---------------------------------------------------------------------------
-// Generic file importer
-// ---------------------------------------------------------------------------
-/**
- * Split a generic file into chunks.
- * Markdown: split on headings and bullet points.
- * Plain text: split on double newlines or one-per-line.
- */
+/** Split a generic file into chunks: markdown on headings and bullets, plain text on double newlines or one per line. */
 function parseGenericFile(filePath) {
     const raw = fs.readFileSync(filePath, 'utf8');
     const isMarkdown = filePath.endsWith('.md') || filePath.endsWith('.mdx');

package/dist/importers/vault.d.ts

@@ -1,11 +1,5 @@
 import { type ImportResult, type ImportOptions } from './core.js';
-/**
- * Import a markdown vault FOLDER as `kind='raw'` memories.
- *
- * NOT re-entrant: idempotency rests on the in-memory `existing` Map loaded once
- * at the top. Two concurrent importVault runs over the same vault could both see
- * a note as absent and double-insert (the connector pattern relies on a single
- * sequential writer; same caveat applies here).
- */
+/** Import a markdown vault FOLDER as `kind='raw'` memories.
+ *  NOT re-entrant: idempotency rests on the in-memory `existing` Map loaded once, so two concurrent runs over one vault can double-insert. */
 export declare function importVault(folderPath: string, options: ImportOptions): ImportResult;
 //# sourceMappingURL=vault.d.ts.map

package/dist/importers/vault.js

@@ -1,11 +1,12 @@
+import { DEFAULT_TENANT_ID } from '../util/env.js';
 import * as fs from 'fs';
 import * as path from 'path';
 import { createHash } from 'node:crypto';
 import { createMemory } from '../core/memory.js';
 import { loadVaultRawRows } from '../store/entry-reads.js';
 import { remember, archiveRaw, isPrivateScope } from '../api/index.js';
-import { assertClientScope } from '../store/recall-scope.js';
-import { RejectedValueError } from '../store/rejection.js';
+import { assertClientScope } from '../core/recall-scope.js';
+import { RejectedValueError } from '../core/api-errors.js';
 import { rejectionGuardRefuses } from '../store/rejected-values.js';
 import { withRequestStoresSync } from '../db/request-stores.js';
 import { loadConfig } from '../core/config.js';
@@ -15,33 +16,17 @@ import { splitMarkdownFrontmatter, frontmatterList, parseWikilinks, collectMarkd
 import { realpathOrResolve } from '../util/real-path.js';
 import { isJsonString } from '../util/json.js';
 import { escapeLike } from '../util/escape.js';
-/**
- * Import a markdown vault FOLDER as `kind='raw'` memories.
- *
- * NOT re-entrant: idempotency rests on the in-memory `existing` Map loaded once
- * at the top. Two concurrent importVault runs over the same vault could both see
- * a note as absent and double-insert (the connector pattern relies on a single
- * sequential writer; same caveat applies here).
- */
+/** Import a markdown vault FOLDER as `kind='raw'` memories.
+ *  NOT re-entrant: idempotency rests on the in-memory `existing` Map loaded once, so two concurrent runs over one vault can double-insert. */
 export function importVault(folderPath, options) {
     const hippoRoot = options.hippoRoot;
     const identity = vaultIdentityOrThrow(options);
     if (options.global) {
-        // The raw-archive path is tenant-local; global mode would put raw vault rows
-        // in the wrong store. Reject for SDK callers too (the CLI also rejects
-        // --global) rather than silently writing local (codex P2).
+        // The raw-archive path is tenant-local, so global mode would put raw vault rows in the wrong store; reject for SDK callers too, not only the CLI.
         throw new Error('importVault does not support global mode (raw rows are tenant-local).');
     }
-    // Self-store no-op guard (codex R8 P1). MUST run BEFORE the existing-rows load
-    // and the deletion-sync pass below. If the vault folder IS the store (or lives
-    // inside it), there are no real vault notes - only the store's own markdown
-    // mirror files. Letting collectMarkdownFiles return [] for this case is NOT
-    // safe: an empty scan is indistinguishable from "every note was deleted", so
-    // deletion-sync would archive every live vault:<name>:* row, and raw-archive
-    // content redaction makes that loss IRREVERSIBLE. The only safe reading of
-    // "import the store into itself" is "do nothing". Canonicalize both paths
-    // (realpath: dereference junctions/symlinks + normalize Windows case) so an
-    // aliased path to the store is still caught (codex R9 P2).
+    // Self-store no-op guard: MUST run before the existing-rows load and the deletion-sync. If the vault IS (or sits inside) the store, an empty scan looks like
+    // "every note deleted" and deletion-sync would irreversibly archive every live row, so do nothing. Canonicalize both paths (realpath) to catch aliased paths.
     const resolvedStore = realpathOrResolve(hippoRoot);
     const resolvedFolder = realpathOrResolve(folderPath);
     if (resolvedFolder === resolvedStore || resolvedFolder.startsWith(resolvedStore + path.sep)) {
@@ -51,7 +36,7 @@ export function importVault(folderPath, options) {
 }
 function syncVaultFolder(folderPath, options, { vaultName, scope }) {
     const hippoRoot = options.hippoRoot;
-    const tenantId = options.tenantId ?? 'default';
+    const tenantId = options.tenantId ?? DEFAULT_TENANT_ID;
     const extraTags = options.extraTags ?? [];
     const dryRun = options.dryRun ?? false;
     const ctx = {
@@ -78,10 +63,8 @@ function syncVaultFolder(folderPath, options, { vaultName, scope }) {
         withRequestStoresSync(importNotes);
     else
         importNotes();
-    // Deletion-sync: any artifactRef present in the Map but NOT seen this run is a
-    // note that vanished from the source folder → archive its raw row. Per-file
-    // archiveRaw (own handle); no outer SAVEPOINT (no cross-file idempotency row
-    // to commit atomically, unlike github's multi-row case).
+    // Deletion-sync: an artifactRef in the Map but not seen this run vanished from the source, so archive its raw row.
+    // Per-file archiveRaw, no outer SAVEPOINT: there is no cross-file idempotency row to commit atomically.
     for (const [artifactRef, rows] of existing) {
         if (seen.has(artifactRef))
             continue;
@@ -95,35 +78,24 @@ function syncVaultFolder(folderPath, options, { vaultName, scope }) {
     return { total, imported, skipped, rejected, archived, redacted, entries };
 }
 function vaultIdentityOrThrow(options) {
-    // Vault NAME is the identity key for the destructive deletion-sync below, so it
-    // must be explicit. Defaulting to the folder basename meant two unrelated vaults
-    // sharing a basename (e.g. work/notes and personal/notes) collided on the same
-    // `vault:<name>:*` prefix - importing the second loaded the first's rows and the
-    // deletion-sync archived them (codex R10 P2). Require a deliberate name instead
-    // of inferring a path-unstable one.
+    // The vault NAME keys the destructive deletion-sync, so it must be explicit: defaulting to the folder basename made two vaults
+    // sharing a basename collide on the same `vault:<name>:*` prefix and the sync archived the first one's rows.
     const vaultName = options.name?.trim();
     if (!vaultName) {
-        throw new Error('importVault requires an explicit vault name (options.name): it is the identity key for source-deletion sync and must not be inferred from the folder basename.');
+        throw new Error('importVault requires an explicit vault name (options.name): it is the identity key for source-deletion sync and must not be inferred from the folder ' +
+            'basename.');
     }
     if (vaultName.includes(':')) {
-        // ':' is the artifactRef delimiter (vault:<name>:<relpath>); a name
-        // containing it lets a different vault's prefix scan over-match and archive
-        // its rows (codex P2). Reject rather than silently corrupt the keys.
+        // ':' delimits the artifactRef (vault:<name>:<relpath>); a name containing it lets another vault's prefix scan over-match and archive its rows.
         throw new Error(`vault name must not contain ':' (artifactRef delimiter): ${vaultName}`);
     }
     const scope = options.scope ?? null;
-    // Privacy footgun guard (codex R13 P2): hippo's recall filter only default-denies
-    // scopes shaped `<source>:private:*` (see isPrivateScope / PRIVATE_SCOPE_RE in
-    // scope.ts). A bare `private` (or `private:<x>`) first segment is NOT recognized
-    // as private, so notes a user believes are private would still be returned to
-    // no-scope recall callers. Reject the alias and point at the source-prefixed form
-    // rather than silently storing public-visible "private" notes. Use recall's own
-    // isPrivateScope as the single source of truth: reject a scope that names a
-    // `private` segment yet is NOT a valid `<source>:private:*` (catches `private`,
-    // `private:x`, and `vault:private` with a missing trailing segment).
+    // Privacy guard: recall only default-denies scopes shaped `<source>:private:*` (isPrivateScope in src/core/recall-scope.ts), so a bare `private`
+    // alias would be returned to no-scope callers. Reject any scope naming a `private` segment that is not a valid `<source>:private:*`.
     assertClientScope(scope);
     if (scope !== null && scope.split(':').includes('private') && !isPrivateScope(scope)) {
-        throw new Error(`vault scope '${scope}' is not recognized as private by recall (only '<source>:private:*' scopes are default-denied). Use a source-prefixed scope such as 'vault:private:${vaultName}'.`);
+        throw new Error(`vault scope '${scope}' is not recognized as private by recall (only '<source>:private:*' scopes are default-denied). ` +
+            `Use a source-prefixed scope such as 'vault:private:${vaultName}'.`);
     }
     return { vaultName, scope };
 }
@@ -153,27 +125,17 @@ function importVaultNote(run, relpath) {
     tally.total++;
     const artifactRef = `vault:${run.vaultName}:${relpath}`;
     run.seen.add(artifactRef);
-    // Content-hash is computed from the RAW file bytes (deterministic; no
-    // Date/random in the content path) so idempotency survives frontmatter
-    // edits identically to body edits.
+    // The content hash covers the RAW file bytes (deterministic, no Date/random) so frontmatter edits are idempotent like body edits.
     const rawFileContent = readVaultNote(run, relpath);
     if (rawFileContent === null)
         return;
     const hash = createHash('sha256').update(rawFileContent).digest('hex');
     const hashTag = `content-hash:${hash}`;
-    // Load every live raw row for this ref (>1 only after a concurrent double-
-    // insert). The idempotency decision happens below, AFTER the full tag set is
-    // built, so it can compare the complete envelope rather than a subset.
+    // Load every live raw row for this ref (>1 only after a concurrent double-insert); decide idempotency below, once the full tag set is built.
     const priors = run.existing.get(artifactRef) ?? [];
     const { fm, body } = splitMarkdownFrontmatter(rawFileContent);
-    // Empty / frontmatter-only note: nothing storable (createMemory enforces a
-    // min content length). The note's CONTENT was deleted at source, so this is a
-    // content-deletion: archive any prior row(s) - the old body must not stay live
-    // and searchable after the source no longer holds it (codex R12 P2) - then
-    // skip the write. Archiving here is safe precisely because we then skip
-    // remember() entirely: there is no archive-then-throw-on-empty-body hazard
-    // (the original reason this branch did not archive). The note stays in `seen`
-    // so deletion-sync does not double-process it.
+    // Empty or frontmatter-only note: nothing storable, and the content was deleted at source, so archive prior rows (the old body must not stay searchable)
+    // and skip the write. Safe because remember() is skipped, so there is no archive-then-throw hazard; the note stays in `seen` so deletion-sync skips it.
     if (body.trim().length < 3) {
         archivePriors(run, priors, `emptied:${artifactRef}`);
         tally.skipped++;
@@ -189,15 +151,10 @@ function importVaultNote(run, relpath) {
 }
 function writeChangedNote(run, artifactRef, priors, body, tags) {
     const { ctx, tally } = run;
-    // Changed file → archive EVERY old raw row for this ref (normally one; >1
-    // only after a concurrent double-insert), then append the new one. NEVER
-    // supersede (would yield kind='distilled'). archiveRaw commits + closes its
-    // handle before remember() runs, so there is no double-live row; a crash
-    // between them self-heals (file re-imported as fresh raw next run).
+    // Changed file: archive EVERY old raw row for this ref (>1 only after a concurrent double-insert), then append the new one; NEVER supersede.
+    // archiveRaw commits before remember() runs, so no double-live row; a crash between them self-heals on the next import.
     archivePriors(run, priors, `changed:${artifactRef}`);
-    // remember() owns the actual write. We build an `echo` of the SAME content +
-    // tags via createMemory purely for the ImportResult, then reconcile its id to
-    // remember()'s real row id so entries[] reflects the row that landed.
+    // remember() owns the write; the `echo` from createMemory only fills the ImportResult, with its id reconciled to the row that landed.
     const content = vetSecrets(body, tags, true).content;
     const echo = createMemory(content, {
         kind: 'raw',
@@ -236,10 +193,7 @@ function archivePriors(run, priors, reason) {
     }
 }
 function vaultNoteTags(run, hashTag, fm, body) {
-    // Build the FULL tag envelope this import would write, BEFORE the idempotency
-    // decision. De-duplicate (createMemory stores tags verbatim, so a collision
-    // between, e.g., a frontmatter tag and an extraTag would otherwise produce a
-    // duplicate). Order-preserving.
+    // Build the FULL tag envelope before the idempotency decision, de-duplicated and order-preserving (createMemory stores tags verbatim).
     const frontmatterTags = [
         ...frontmatterList(fm['tags']),
         ...frontmatterList(fm['aliases']).map((a) => `alias:${a}`),
@@ -255,14 +209,8 @@ function vaultNoteTags(run, hashTag, fm, body) {
     ]));
 }
 function vaultEnvelopeUnchanged(priors, tags, scope) {
-    // Unchanged iff EVERY live raw row carries the EXACT same tag set AND scope.
-    // Comparing the COMPLETE set (not the content-hash + a subset of extra tags)
-    // means every envelope change registers: content (via the content-hash tag),
-    // frontmatter, wikilinks, an ADDED extra tag, or a REMOVED one - the earlier
-    // piecemeal checks missed scope (R10 P2) then tag removal (R11 P2). Set
-    // equality is order-independent and both sides are deduped. (`length > 0`
-    // guard: a never-seen file must import, not skip; archiving ALL priors on a
-    // mismatch also clears any concurrent-double-insert duplicates.)
+    // Unchanged iff EVERY live raw row carries the EXACT same tag set AND scope; comparing the complete envelope (not hash plus a subset) catches
+    // scope, added and removed tags. `length > 0` guards a never-seen file; archiving all priors on a mismatch also clears concurrent double-inserts.
     const wantTags = new Set(tags);
     return (priors.length > 0 &&
         priors.every((p) => {
@@ -297,9 +245,7 @@ function storeVaultNote(run, echo, content, tags, artifactRef) {
         throw err;
     }
 }
-/** Local tolerant JSON-array parse for the loader's `tags_json` column. The
- *  store's own `parseJsonArray` is not exported; this matches its contract
- *  (returns [] on null/garbage). */
+/** Tolerant JSON-array parse for the loader's `tags_json` column (the store's `parseJsonArray` is not exported); returns [] on null or garbage. */
 function parseJsonArrayLoose(value) {
     if (!value)
         return [];

package/dist/index.d.ts

@@ -32,10 +32,15 @@ export { consolidate } from './consolidate/sleep.js';
 export { ConsolidationResult } from './consolidate/run.js';
 export { sleep, type SleepOpts, type SleepResult } from './api/sleep.js';
 export { strengthBucket } from './consolidate/dedupe.js';
-export { isEmbeddingAvailable, getEmbedding } from './store/embeddings/local.js';
-export { cosineSimilarity, loadEmbeddingIndex, saveEmbeddingIndex, embedMemory, embedAll, } from './store/embeddings/index.js';
+export { isEmbeddingAvailable, getEmbedding } from './embeddings/local.js';
+export { cosineSimilarity } from './store/embeddings/index.js';
+export { embedMemory, embedAll } from './search/embed-memories.js';
+export { loadEmbeddingIndex, saveEmbeddingIndex } from './store/vector-index.js';
 export { captureError, extractLessons, partitionLessons, deduplicateLesson, runWatched, fetchGitLog, } from './learn/autolearn.js';
-export { getGlobalRoot, initGlobal, promoteToGlobal, searchBoth, searchBothHybrid, syncGlobalToLocal, HybridSearchOptions, transferScore, shareMemory, listPeers, autoShare, } from './sharing/shared.js';
+export { getGlobalRoot, initGlobal, promoteToGlobal } from './sharing/global-store.js';
+export { searchBoth, searchBothHybrid, HybridSearchOptions } from './sharing/search-both.js';
+export { syncGlobalToLocal } from './sharing/global-sync.js';
+export { transferScore, shareMemory, listPeers, autoShare } from './sharing/share.js';
 export { wmPush, wmRead, wmClear, wmFlush, WorkingMemoryItem, WM_MAX_ENTRIES, } from './store/working-memory.js';
 export { importChatGPT, importClaude, importCursor, importGenericFile, } from './importers/sources.js';
 export { importMarkdown } from './importers/markdown.js';

package/dist/index.js

@@ -36,12 +36,17 @@ export { sleep } from './api/sleep.js';
 // Announced public in CHANGELOG 1.26.3 but never re-exported; the rest of dedupe.js stays internal.
 export { strengthBucket } from './consolidate/dedupe.js';
 // Feature 1: Embedding search
-export { isEmbeddingAvailable, getEmbedding } from './store/embeddings/local.js';
-export { cosineSimilarity, loadEmbeddingIndex, saveEmbeddingIndex, embedMemory, embedAll, } from './store/embeddings/index.js';
+export { isEmbeddingAvailable, getEmbedding } from './embeddings/local.js';
+export { cosineSimilarity } from './store/embeddings/index.js';
+export { embedMemory, embedAll } from './search/embed-memories.js';
+export { loadEmbeddingIndex, saveEmbeddingIndex } from './store/vector-index.js';
 // Feature 2: Auto-learn from errors
 export { captureError, extractLessons, partitionLessons, deduplicateLesson, runWatched, fetchGitLog, } from './learn/autolearn.js';
 // Feature 3: Cross-agent shared memory
-export { getGlobalRoot, initGlobal, promoteToGlobal, searchBoth, searchBothHybrid, syncGlobalToLocal, transferScore, shareMemory, listPeers, autoShare, } from './sharing/shared.js';
+export { getGlobalRoot, initGlobal, promoteToGlobal } from './sharing/global-store.js';
+export { searchBoth, searchBothHybrid } from './sharing/search-both.js';
+export { syncGlobalToLocal } from './sharing/global-sync.js';
+export { transferScore, shareMemory, listPeers, autoShare } from './sharing/share.js';
 // Feature 5: Working memory
 export { wmPush, wmRead, wmClear, wmFlush, WM_MAX_ENTRIES, } from './store/working-memory.js';
 // Feature 4: Memory importers

package/dist/learn/autolearn.d.ts

@@ -1,56 +1,37 @@
-/**
- * Auto-learn from errors and git history.
- * Agents learn from failures without explicit hippo remember calls.
- */
+/** Auto-learn from errors and git history.
+ *  Agents learn from failures without explicit hippo remember calls. */
 import { MemoryEntry } from '../core/memory.js';
-/** A memory of a failed command, "Command '<cmd>' failed: <truncated stderr>"; no store is in reach, so `hippo watch` re-derives its half-life from the store's config. */
+/** A memory of a failed command, "Command '<cmd>' failed: <truncated stderr>"; `hippo watch` stores its text and tags through remember. */
 export declare function captureError(exitCode: number, stderr: string, command: string, tenantId?: string): MemoryEntry;
-/**
- * Parse git log output for actionable lessons.
- * Looks for fix:, revert:, bug:, error:, hotfix: commit messages.
- */
+/** Parse git log output for actionable lessons.
+ *  Looks for fix:, revert:, bug:, error:, hotfix: commit messages. */
 export declare function extractLessons(gitLog: string, customPatterns?: string[]): string[];
-/**
- * Split parsed lessons into ones worth storing and low-information ones.
- *
- * `extractLessons` is a published API surface (exported from src/index.ts)
- * and does one job: parse a git log into candidate lesson strings. Adding a
- * quality check inside it would break that contract and silently change
- * what external callers get back. So parsing and admission are separate
- * steps: this function is the write-path gate, called by each caller that
- * actually stores a lesson, not by the parser itself.
- *
- * Order is preserved in both output arrays; both hold lessons with secret shapes redacted.
- */
+/** Split parsed lessons into ones worth storing and low-information ones; kept out of `extractLessons` so that published parser's output cannot change.
+ *  Order is preserved in both arrays, and both hold lessons with secret shapes redacted. */
 export declare function partitionLessons(lessons: string[]): {
     kept: string[];
     dropped: string[];
 };
-/**
- * Check if a substantially similar memory already exists.
- * Returns true if overlap > threshold (default 0.7).
- *
- * `tenantId` is opt-in. Only takes effect when the first argument is a
- * root string (string-overload path). When the first argument is a
- * pre-loaded MemoryEntry[], the caller has already scoped — tenantId is
- * ignored on that path.
- */
+/** Check if a substantially similar memory already exists: true if overlap > threshold (default 0.7).
+ *  `tenantId` applies only on the root-string overload; a pre-loaded MemoryEntry[] is already scoped, so it is ignored there. */
 export declare function deduplicateLesson(hippoRootOrEntries: string | MemoryEntry[], lesson: string, threshold?: number, tenantId?: string): boolean;
-/**
- * Run a command in a shell, streaming stderr live; never pass text the caller did not write.
- * Returns: { exitCode, stderr }.
- */
-export declare function runWatched(command: string): Promise<{
+/** A transfer command that has not exited in this long is hung on a prompt or a dead remote. */
+export declare const RUN_WATCHED_DEADLINE_MS: number;
+/** The command passed its deadline and was killed. */
+export declare class RunWatchedTimeoutError extends Error {
+    readonly deadlineMs: number;
+    readonly stderr: string;
+    constructor(deadlineMs: number, stderr: string);
+}
+/** Run a command in a shell, streaming stderr live; never pass text the caller did not write.
+ *  Returns: { exitCode, stderr } with only the last 64 KiB of stderr; rejects with RunWatchedTimeoutError past `deadlineMs`. */
+export declare function runWatched(command: string, deadlineMs?: number): Promise<{
     exitCode: number;
     stderr: string;
 }>;
-/**
- * Check whether a directory is a git work tree.
- */
+/** Check whether a directory is a git work tree. */
 export declare function isGitRepo(cwd: string): boolean;
-/**
- * Fetch recent git log lines (subject lines only).
- * days: how many days of history to include.
- */
+/** Fetch recent git log lines (subject lines only).
+ *  days: how many days of history to include. */
 export declare function fetchGitLog(cwd: string, days: number): string;
 //# sourceMappingURL=autolearn.d.ts.map