bikky 0.4.1 → 0.4.3

package/dist/daemon/session-index.js

@@ -8,6 +8,7 @@ import { createHash, randomUUID } from "node:crypto";
 import { CAPTURE_POLICY_VERSION, DEFAULT_CAPTURE_CONTEXT, PROMPT_VERSIONS } from "./capture-policy.js";
 import * as qdrant from "./qdrant.js";
 import { combineRedactions, redactStorageText, } from "../privacy/redaction.js";
+import { buildOperationOrigin } from "../provenance/origin.js";
 const contentHash = (text) => createHash("sha256").update(`session-index:${text}`).digest("hex");
 export const buildSessionIndexDraft = (input) => {
     const episodeIds = input.episodeResults
@@ -48,6 +49,16 @@ export const buildSessionIndexPayload = (input) => {
     const redactedEntities = input.draft.entities.map((entity) => redactStorageText(entity, input.redactionOptions));
     const redaction = combineRedactions([redactedContent, ...redactedEntities]);
     const existingPayload = input.existing?.payload ?? {};
+    const operationOrigin = input.origin ?? buildOperationOrigin({
+        interface: "daemon",
+        action: input.existing ? "update" : "create",
+        subsystem: "session_index",
+        config: input.config,
+        metadata: {
+            session_id: input.sessionId,
+            event_count: input.eventCount,
+        },
+    });
     const payload = {
         ...existingPayload,
         content: redactedContent.text,
@@ -57,9 +68,9 @@ export const buildSessionIndexPayload = (input) => {
         memory_subtype: "session_index",
         layer: "episode",
         ...(input.scope.workspaceId ? { workspace_id: input.scope.workspaceId } : {}),
-        ...(input.scope.actorId ? { actor_id: input.scope.actorId } : {}),
+        origin: existingPayload.origin ?? operationOrigin,
+        ...(input.existing ? { last_operation_origin: operationOrigin } : {}),
         entities: redactedEntities.map((entity) => entity.text.toLowerCase()),
-        source: "system",
         confidence: 1.0,
         importance: input.draft.importance,
         content_hash: contentHash(redactedContent.text),
@@ -93,24 +104,38 @@ export const buildSessionIndexPayload = (input) => {
     }
     return { payload, redaction };
 };
-const findExistingSessionIndex = async (sessionId, scope) => {
-    const result = await qdrant.qdrantRequest("POST", `/collections/${qdrant.collection}/points/scroll`, {
+const findExistingSessionIndex = async (sessionId, scope, destination) => {
+    const collection = qdrant.collectionForDestination(destination);
+    const result = await qdrant.qdrantRequest("POST", `/collections/${collection}/points/scroll`, {
         filter: buildSessionIndexFilter(sessionId, scope),
         limit: 1,
         with_payload: true,
-    });
+    }, destination);
     return result.result?.points?.[0] ?? null;
 };
 export const updateSessionIndex = async (input) => {
     if (input.episodeResults.length === 0) {
         return { action: "skipped", reason: "no_episode_results" };
     }
-    const existing = await findExistingSessionIndex(input.sessionId, input.scope);
     const draft = buildSessionIndexDraft({
         sessionId: input.sessionId,
         eventCount: input.eventCount,
         episodeResults: input.episodeResults,
     });
+    const destination = input.destination
+        ?? input.episodeResults.find((result) => result.destination)?.destination
+        ?? qdrant.resolveDestination({
+            content: draft.content,
+            entities: draft.entities,
+            metadata: {
+                session_id: input.sessionId,
+                memory_subtype: "session_index",
+                kind: "summary",
+                origin_interface: "daemon",
+                origin_agent_type: "daemon",
+            },
+        }).name;
+    const existing = await findExistingSessionIndex(input.sessionId, input.scope, destination);
     const now = new Date().toISOString();
     const { payload } = buildSessionIndexPayload({
         draft,
@@ -119,6 +144,7 @@ export const updateSessionIndex = async (input) => {
         now,
         existing,
         eventCount: input.eventCount,
+        config: input.config,
         redactionOptions: {
             enabled: true,
             redactPii: false,
@@ -126,9 +152,10 @@ export const updateSessionIndex = async (input) => {
     });
     const vector = await qdrant.embed(String(payload.content));
     const factId = existing?.id ?? randomUUID();
-    await qdrant.qdrantRequest("PUT", `/collections/${qdrant.collection}/points`, {
+    const collection = qdrant.collectionForDestination(destination);
+    await qdrant.qdrantRequest("PUT", `/collections/${collection}/points`, {
         points: [{ id: factId, vector, payload }],
-    });
-    return { action: existing ? "updated" : "stored", factId };
+    }, destination);
+    return { action: existing ? "updated" : "stored", factId, destination };
 };
 //# sourceMappingURL=session-index.js.map

package/dist/daemon/session-summary.d.ts

@@ -1,6 +1,7 @@
 import type { BikkyConfig } from "../config.js";
 import type { QdrantPayload } from "./qdrant.js";
 import { type RedactionSummary } from "../privacy/redaction.js";
+import { type OperationOrigin } from "../provenance/origin.js";
 export interface WorkspaceScope {
     workspaceId?: string;
     actorId?: string;
@@ -52,6 +53,8 @@ export declare const buildSessionSummaryPayload: (input: {
         enabled: boolean;
         redactPii: boolean;
     };
+    config?: BikkyConfig;
+    origin?: OperationOrigin;
 }) => SessionSummaryPayloadResult;
 export declare const updateSessionSummary: (input: {
     sessionId: string;

package/dist/daemon/session-summary.js

@@ -13,7 +13,7 @@ import { buildSessionIndexFilter, updateSessionIndex } from "./session-index.js"
 import { updateWorkstreamSummaries } from "./workstream-summary.js";
 import * as qdrant from "./qdrant.js";
 import { combineRedactions, redactStorageText, } from "../privacy/redaction.js";
-import { resolveActorIdentity } from "../provenance/actor.js";
+import { buildOperationOrigin } from "../provenance/origin.js";
 const DEFAULT_SUMMARY_IMPORTANCE = 0.8;
 const contentHash = (text) => createHash("sha256").update(`summary:${text}`).digest("hex");
 const stripJsonFence = (raw) => raw.trim().replace(/^```(?:json)?\s*\n?/, "").replace(/\n?```\s*$/, "").trim();
@@ -92,6 +92,16 @@ export const buildSessionSummaryPayload = (input) => {
         ...redactedEntities,
     ]);
     const existingPayload = input.existing?.payload ?? {};
+    const operationOrigin = input.origin ?? buildOperationOrigin({
+        interface: "daemon",
+        action: input.existing ? "update" : "create",
+        subsystem: "session_summary",
+        config: input.config,
+        metadata: {
+            session_id: input.sessionId,
+            event_count: input.eventCount,
+        },
+    });
     const payload = {
         ...existingPayload,
         content: redactedContent.text,
@@ -101,9 +111,9 @@ export const buildSessionSummaryPayload = (input) => {
         memory_subtype: "session_index",
         layer: "episode",
         ...(input.scope.workspaceId ? { workspace_id: input.scope.workspaceId } : {}),
-        ...(input.scope.actorId ? { actor_id: input.scope.actorId } : {}),
+        origin: existingPayload.origin ?? operationOrigin,
+        ...(input.existing ? { last_operation_origin: operationOrigin } : {}),
         entities: redactedEntities.map((entity) => entity.text.toLowerCase()),
-        source: "system",
         confidence: 1.0,
         importance: input.draft.importance,
         content_hash: contentHash(redactedContent.text),
@@ -144,8 +154,7 @@ export const updateSessionSummary = async (input) => {
         return { action: "skipped", reason: "qdrant_not_ready" };
     }
     const config = input.config ?? loadConfig();
-    const actor = resolveActorIdentity({ config });
-    const scope = actor.actor_id ? { actorId: actor.actor_id } : {};
+    const scope = {};
     const segments = segmentTranscriptIntoEpisodes({
         sessionId: input.sessionId,
         transcript: input.transcript,
@@ -164,18 +173,42 @@ export const updateSessionSummary = async (input) => {
     if (episodeResults.length === 0) {
         return { action: "skipped", reason: "no_episode_summaries" };
     }
-    const indexResult = await updateSessionIndex({
-        sessionId: input.sessionId,
-        eventCount: input.eventCount,
-        episodeResults,
-        scope,
-        config,
-    });
+    let indexResult = null;
+    const episodeResultsByDestination = new Map();
+    for (const result of episodeResults) {
+        const destination = result.destination ?? qdrant.resolveDestination({
+            content: result.workstreamKey ?? result.episodeId ?? input.sessionId,
+            entities: [result.workstreamKey, result.episodeId].filter((value) => Boolean(value)),
+            metadata: {
+                session_id: input.sessionId,
+                ...(result.episodeId ? { episode_id: result.episodeId } : {}),
+                ...(result.workstreamKey ? { workstream_key: result.workstreamKey } : {}),
+                kind: "summary",
+                memory_subtype: "session_index",
+                origin_interface: "daemon",
+                origin_agent_type: "daemon",
+            },
+        }).name;
+        const bucket = episodeResultsByDestination.get(destination) ?? [];
+        bucket.push(result);
+        episodeResultsByDestination.set(destination, bucket);
+    }
+    for (const [destination, destinationEpisodeResults] of episodeResultsByDestination) {
+        const result = await updateSessionIndex({
+            sessionId: input.sessionId,
+            eventCount: input.eventCount,
+            episodeResults: destinationEpisodeResults,
+            scope,
+            config,
+            destination,
+        });
+        indexResult ??= result;
+    }
     await updateWorkstreamSummaries({ episodeResults, scope, config });
     return {
-        action: indexResult.action,
-        factId: indexResult.factId,
-        reason: indexResult.reason,
+        action: indexResult?.action ?? "skipped",
+        factId: indexResult?.factId,
+        reason: indexResult?.reason,
     };
 };
 //# sourceMappingURL=session-summary.js.map

package/dist/daemon/staleness.js

@@ -11,7 +11,7 @@ let logFn = console.log;
 let lastStaleIds = "";
 const defaultDeps = {
     isReady: qdrantMod.isReady,
-    scrollFacts: qdrantMod.scrollFacts,
+    scrollFacts: qdrantMod.scrollFactsAcrossDestinations,
 };
 /**
  * Scan for stale facts via direct Qdrant scroll query.
@@ -37,7 +37,7 @@ export const scanStaleFacts = async (config, deps = defaultDeps) => {
         }, limit);
         if (staleFacts.length > 0) {
             // Deduplicate: only log if the set of stale fact IDs has changed
-            const currentIds = staleFacts.map((f) => f.id).sort().join(",");
+            const currentIds = staleFacts.map((f) => `${f.destination ?? "default"}:${f.id}`).sort().join(",");
             if (currentIds === lastStaleIds) {
                 logFn("DEBUG", `Staleness scan: same ${staleFacts.length} fact(s) still stale, skipping duplicate log`);
                 return;

package/dist/daemon/transcript-sources.js

@@ -166,11 +166,12 @@ export const parseClaudeTranscriptLine = (line) => {
 };
 export const readNewTranscriptEvents = async (eventsPath, byteOffset, source) => {
     const fileStat = await stat(eventsPath);
-    if (fileStat.size <= byteOffset) {
+    if (fileStat.size === byteOffset) {
         return { events: [], newOffset: byteOffset, totalLines: 0 };
     }
+    const startOffset = fileStat.size < byteOffset ? 0 : byteOffset;
     const buf = await readFile(eventsPath);
-    const newContent = buf.subarray(byteOffset).toString("utf-8");
+    const newContent = buf.subarray(startOffset).toString("utf-8");
     const events = [];
     let totalLines = 0;
     for (const line of newContent.split("\n")) {

package/dist/daemon/watcher.js

@@ -6,6 +6,8 @@ import path from "node:path";
 import { loadConfig } from "../config.js";
 export function discoverSessions() {
     const cfg = loadConfig();
+    if (!cfg.watchers.copilot.enabled)
+        return [];
     const baseDir = cfg.watchers.copilot.path;
     if (!fs.existsSync(baseDir))
         return [];

package/dist/daemon/workstream-summary.d.ts

@@ -2,6 +2,7 @@ import type { BikkyConfig } from "../config.js";
 import type { EpisodeSummaryWriteResult } from "./episode-summary.js";
 import type { QdrantPayload } from "./qdrant.js";
 import { type RedactionSummary } from "../privacy/redaction.js";
+import { type OperationOrigin } from "../provenance/origin.js";
 export { buildWorkstreamSummaryMessages } from "../prompts/index.js";
 export interface WorkspaceScope {
     workspaceId?: string;
@@ -23,6 +24,7 @@ export interface WorkstreamSummaryPayloadResult {
 export interface WorkstreamUpdateResult {
     action: "stored" | "updated" | "skipped";
     factId?: string;
+    destination?: string;
     workstreamKey?: string;
     reason?: string;
 }
@@ -45,6 +47,8 @@ export declare const buildWorkstreamSummaryPayload: (input: {
         enabled: boolean;
         redactPii: boolean;
     };
+    config?: BikkyConfig;
+    origin?: OperationOrigin;
 }) => WorkstreamSummaryPayloadResult;
 export declare const updateWorkstreamSummaries: (input: {
     episodeResults: EpisodeSummaryWriteResult[];

package/dist/daemon/workstream-summary.js

@@ -11,6 +11,7 @@ import { workstreamSummaryPrompt } from "../prompts/index.js";
 import { CAPTURE_POLICY_VERSION, CAPTURE_TRIGGERS, DEFAULT_CAPTURE_CONTEXT, PROMPT_VERSIONS, } from "./capture-policy.js";
 import * as qdrant from "./qdrant.js";
 import { combineRedactions, redactStorageText, } from "../privacy/redaction.js";
+import { buildOperationOrigin } from "../provenance/origin.js";
 export { buildWorkstreamSummaryMessages } from "../prompts/index.js";
 const contentHash = (text) => createHash("sha256").update(`workstream:${text}`).digest("hex");
 const stripJsonFence = (raw) => raw.trim().replace(/^```(?:json)?\s*\n?/, "").replace(/\n?```\s*$/, "").trim();
@@ -39,6 +40,20 @@ export const groupEpisodeResultsByWorkstream = (episodeResults) => {
     }
     return grouped;
 };
+const groupEpisodeResultsByWorkstreamDestination = (episodeResults) => {
+    const grouped = new Map();
+    for (const result of episodeResults) {
+        const workstreamKey = result.workstreamKey?.trim();
+        if (!workstreamKey || !result.episodeId)
+            continue;
+        const destination = result.destination?.trim();
+        const groupKey = `${destination ?? ""}::${workstreamKey}`;
+        const existing = grouped.get(groupKey) ?? { workstreamKey, destination, results: [] };
+        existing.results.push(result);
+        grouped.set(groupKey, existing);
+    }
+    return grouped;
+};
 export const parseWorkstreamSummaryDraft = (raw) => {
     const parsed = JSON.parse(stripJsonFence(raw));
     const contentValue = parsed.content ?? parsed.summary;
@@ -109,6 +124,17 @@ export const buildWorkstreamSummaryPayload = (input) => {
         ...redactedEntities,
     ]);
     const existingPayload = input.existing?.payload ?? {};
+    const operationOrigin = input.origin ?? buildOperationOrigin({
+        interface: "daemon",
+        action: input.existing ? "update" : "create",
+        subsystem: "workstream_summary",
+        config: input.config,
+        metadata: {
+            workstream_key: input.workstreamKey,
+            source_episode_count: input.sourceEpisodeIds.length,
+            ...(input.repo ? { repo: input.repo } : {}),
+        },
+    });
     const payload = {
         ...existingPayload,
         content: redactedContent.text,
@@ -118,10 +144,10 @@ export const buildWorkstreamSummaryPayload = (input) => {
         memory_subtype: "workstream",
         layer: "workstream",
         ...(input.scope.workspaceId ? { workspace_id: input.scope.workspaceId } : {}),
-        ...(input.scope.actorId ? { actor_id: input.scope.actorId } : {}),
         ...(input.repo ? { repo: input.repo } : {}),
+        origin: existingPayload.origin ?? operationOrigin,
+        ...(input.existing ? { last_operation_origin: operationOrigin } : {}),
         entities: redactedEntities.map((entity) => entity.text.toLowerCase()),
-        source: "system",
         confidence: 1.0,
         importance: input.draft.importance,
         content_hash: contentHash(redactedContent.text),
@@ -156,20 +182,22 @@ export const buildWorkstreamSummaryPayload = (input) => {
     }
     return { payload, redaction };
 };
-const findExistingWorkstreamSummary = async (workstreamKey, scope, repo) => {
-    const result = await qdrant.qdrantRequest("POST", `/collections/${qdrant.collection}/points/scroll`, {
+const findExistingWorkstreamSummary = async (workstreamKey, scope, repo, destination) => {
+    const collection = qdrant.collectionForDestination(destination);
+    const result = await qdrant.qdrantRequest("POST", `/collections/${collection}/points/scroll`, {
         filter: buildWorkstreamSummaryFilter(workstreamKey, scope, repo),
         limit: 1,
         with_payload: true,
-    });
+    }, destination);
     return result.result?.points?.[0] ?? null;
 };
-const loadEpisodeSummaries = async (workstreamKey, scope) => {
-    const result = await qdrant.qdrantRequest("POST", `/collections/${qdrant.collection}/points/scroll`, {
+const loadEpisodeSummaries = async (workstreamKey, scope, destination) => {
+    const collection = qdrant.collectionForDestination(destination);
+    const result = await qdrant.qdrantRequest("POST", `/collections/${collection}/points/scroll`, {
         filter: buildWorkstreamEpisodeFilter(workstreamKey, scope),
         limit: CAPTURE_TRIGGERS.workstreamSummary.maxEpisodesPerUpdate,
         with_payload: true,
-    });
+    }, destination);
     return result.result?.points ?? [];
 };
 const summarizeWorkstream = async (input) => {
@@ -183,20 +211,32 @@ const summarizeWorkstream = async (input) => {
     return parseWorkstreamSummaryDraft(result);
 };
 export const updateWorkstreamSummaries = async (input) => {
-    const grouped = groupEpisodeResultsByWorkstream(input.episodeResults);
+    const grouped = groupEpisodeResultsByWorkstreamDestination(input.episodeResults);
     if (grouped.size === 0)
         return [{ action: "skipped", reason: "no_workstream_keys" }];
     const results = [];
-    for (const [workstreamKey] of grouped) {
-        const episodes = await loadEpisodeSummaries(workstreamKey, input.scope);
+    for (const { workstreamKey, destination: groupDestination } of grouped.values()) {
+        const destination = groupDestination
+            ?? qdrant.resolveDestination({
+                content: workstreamKey,
+                entities: [workstreamKey],
+                metadata: {
+                    workstream_key: workstreamKey,
+                    memory_subtype: "workstream",
+                    kind: "summary",
+                    origin_interface: "daemon",
+                    origin_agent_type: "daemon",
+                },
+            }).name;
+        const episodes = await loadEpisodeSummaries(workstreamKey, input.scope, destination);
         const episodeSummaries = episodes
             .map((episode) => episode.payload?.content)
             .filter((content) => Boolean(content?.trim()));
         if (episodeSummaries.length < CAPTURE_TRIGGERS.workstreamSummary.minEpisodeCount) {
-            results.push({ action: "skipped", reason: "not_enough_episodes", workstreamKey });
+            results.push({ action: "skipped", reason: "not_enough_episodes", workstreamKey, destination });
             continue;
         }
-        const existing = await findExistingWorkstreamSummary(workstreamKey, input.scope, null);
+        const existing = await findExistingWorkstreamSummary(workstreamKey, input.scope, null, destination);
         const draft = await summarizeWorkstream({
             workstreamKey,
             existingSummary: existing?.payload?.content ?? null,
@@ -211,6 +251,7 @@ export const updateWorkstreamSummaries = async (input) => {
             existing,
             sourceEpisodeIds: episodes.map((episode) => episode.payload?.episode_id ?? episode.id).filter(Boolean),
             repo: null,
+            config: input.config,
             redactionOptions: {
                 enabled: true,
                 redactPii: false,
@@ -218,10 +259,11 @@ export const updateWorkstreamSummaries = async (input) => {
         });
         const vector = await qdrant.embed(String(payload.content));
         const factId = existing?.id ?? randomUUID();
-        await qdrant.qdrantRequest("PUT", `/collections/${qdrant.collection}/points`, {
+        const collection = qdrant.collectionForDestination(destination);
+        await qdrant.qdrantRequest("PUT", `/collections/${collection}/points`, {
             points: [{ id: factId, vector, payload }],
-        });
-        results.push({ action: existing ? "updated" : "stored", factId, workstreamKey });
+        }, destination);
+        results.push({ action: existing ? "updated" : "stored", factId, workstreamKey, destination });
     }
     return results;
 };

package/dist/install.d.ts

@@ -1,6 +1,7 @@
 /**
  * Write MCP config entries for Copilot and/or Claude Code.
  */
+import { type OriginIdentity } from "./provenance/origin.js";
 export interface InstallOptions {
     homeDir?: string;
     /**
@@ -8,6 +9,16 @@ export interface InstallOptions {
      * user config file directly.
      */
     claudeCommand?: string | null;
+    /**
+     * Defaults to true. Set to false in tests or advanced flows that only want
+     * MCP config files and do not want ~/.bikky/config.json touched.
+     */
+    provisionIdentity?: boolean;
+    env?: NodeJS.ProcessEnv;
+    cwd?: string;
+    hostname?: string;
+    shellUsername?: string | null;
 }
+export declare function provisionUserIdentityConfig(options?: InstallOptions): OriginIdentity | null;
 export declare function writeInstallConfig(options?: InstallOptions): Promise<void>;
 //# sourceMappingURL=install.d.ts.map

package/dist/install.js

@@ -5,6 +5,7 @@ import fs from "node:fs";
 import path from "node:path";
 import os from "node:os";
 import { spawnSync } from "node:child_process";
+import { inferUserIdentity } from "./provenance/origin.js";
 const SERVER_NAME = "bikky";
 function readJsonConfig(filePath) {
     if (!fs.existsSync(filePath))
@@ -57,6 +58,40 @@ function writeClaudeCodeUserConfig(homeDir, entry) {
     writeJsonConfig(claudeConfigPath, config);
     console.log(`✅ Written to ${claudeConfigPath}`);
 }
+function bikkyConfigPath(homeDir, explicitHomeDir, env = process.env) {
+    const bikkyDir = explicitHomeDir ? path.join(homeDir, ".bikky") : env.BIKKY_HOME ?? path.join(homeDir, ".bikky");
+    return path.join(bikkyDir, "config.json");
+}
+function hasConfiguredValue(value) {
+    return typeof value === "string" && value.trim().length > 0;
+}
+export function provisionUserIdentityConfig(options = {}) {
+    const homeDir = options.homeDir ?? os.homedir();
+    const configPath = bikkyConfigPath(homeDir, options.homeDir !== undefined, options.env);
+    const config = readJsonConfig(configPath);
+    const existingIdentity = config.identity && typeof config.identity === "object" && !Array.isArray(config.identity)
+        ? config.identity
+        : {};
+    const hasUserId = hasConfiguredValue(existingIdentity.user_id);
+    const hasUserName = hasConfiguredValue(existingIdentity.user_name);
+    if (hasUserId && hasUserName)
+        return null;
+    const inferred = inferUserIdentity({
+        env: options.env,
+        cwd: options.cwd,
+        hostname: options.hostname,
+        shellUsername: options.shellUsername,
+    });
+    const nextIdentity = { ...existingIdentity };
+    if (!hasUserId)
+        nextIdentity.user_id = inferred.id;
+    if (!hasUserName)
+        nextIdentity.user_name = inferred.name;
+    config.identity = nextIdentity;
+    writeJsonConfig(configPath, config);
+    console.log(`✅ Provisioned bikky user identity in ${configPath}`);
+    return inferred;
+}
 export async function writeInstallConfig(options = {}) {
     const homeDir = options.homeDir ?? os.homedir();
     const entry = {
@@ -72,6 +107,9 @@ export async function writeInstallConfig(options = {}) {
     if (!registeredWithClaudeCli) {
         writeClaudeCodeUserConfig(homeDir, entry);
     }
+    if (options.provisionIdentity !== false) {
+        provisionUserIdentityConfig(options);
+    }
     console.log("\n🧠 bikky is now registered. Restart your editor to activate.");
 }
 //# sourceMappingURL=install.js.map

package/dist/llm/embedding/index.js

@@ -19,6 +19,7 @@
 import "./providers/index.js";
 import { getEmbeddingProvider } from "./registry.js";
 import { EmbeddingDimensionMismatchError } from "../errors.js";
+import { firstNonEmptyString } from "../util.js";
 export { registerEmbeddingProvider, getEmbeddingProvider, listEmbeddingProviders, } from "./registry.js";
 const DEFAULT_TIMEOUT_MS = 30_000;
 const DEFAULT_RETRIES = 2;
@@ -32,7 +33,7 @@ export function initEmbedding(input) {
         provider: provider.name,
         model: input.model ?? provider.defaults.model,
         dimensions: input.dimensions ?? provider.defaults.dimensions,
-        baseUrl: (input.baseUrl ?? provider.defaults.baseUrl ?? "").replace(/\/+$/, ""),
+        baseUrl: (firstNonEmptyString(input.baseUrl, provider.defaults.baseUrl) ?? "").replace(/\/+$/, ""),
         apiKey: input.apiKey ?? null,
         extra: input.extra ?? {},
         timeoutMs: input.timeoutMs ?? DEFAULT_TIMEOUT_MS,

package/dist/llm/embedding/providers/openai.js

@@ -10,12 +10,18 @@ export const openaiEmbeddingProvider = {
     browserCompatible: true,
     defaults: {
         model: "text-embedding-3-small",
-        dimensions: 1536,
+        dimensions: 1024,
         baseUrl: "https://api.openai.com",
     },
     async embed(text, cfg) {
         if (!cfg.apiKey)
             throw new Error("Embedding failed [openai]: api key not configured");
+        const body = { model: cfg.model, input: text };
+        // The `dimensions` parameter (Matryoshka truncation) is only supported by
+        // the text-embedding-3-* family. Older models (e.g. ada-002) reject it.
+        if (cfg.dimensions && /text-embedding-3/.test(cfg.model)) {
+            body.dimensions = cfg.dimensions;
+        }
         const resp = await resilientFetch({
             url: `${cfg.baseUrl}/v1/embeddings`,
             init: {
@@ -24,7 +30,7 @@ export const openaiEmbeddingProvider = {
                     "Content-Type": "application/json",
                     Authorization: `Bearer ${cfg.apiKey}`,
                 },
-                body: JSON.stringify({ model: cfg.model, input: text }),
+                body: JSON.stringify(body),
             },
             timeoutMs: cfg.timeoutMs,
             retries: cfg.retries,

package/dist/llm/embedding/providers/portkey.js

@@ -19,7 +19,7 @@ export const portkeyEmbeddingProvider = {
     browserCompatible: true,
     defaults: {
         model: "@openai/text-embedding-3-small",
-        dimensions: 1536,
+        dimensions: 1024,
         baseUrl: "https://api.portkey.ai",
     },
     async embed(text, cfg) {
@@ -33,12 +33,19 @@ export const portkeyEmbeddingProvider = {
             headers["x-portkey-virtual-key"] = cfg.extra.virtual_key;
         if (cfg.extra.config_id)
             headers["x-portkey-config"] = cfg.extra.config_id;
+        const body = { model: cfg.model, input: text };
+        // The `dimensions` parameter (Matryoshka truncation) is supported by the
+        // OpenAI text-embedding-3-* family. Forward it when the model name matches
+        // so users get the configured vector size instead of the provider native.
+        if (cfg.dimensions && /text-embedding-3/.test(cfg.model)) {
+            body.dimensions = cfg.dimensions;
+        }
         const resp = await resilientFetch({
             url: `${cfg.baseUrl}/v1/embeddings`,
             init: {
                 method: "POST",
                 headers,
-                body: JSON.stringify({ model: cfg.model, input: text }),
+                body: JSON.stringify(body),
             },
             timeoutMs: cfg.timeoutMs,
             retries: cfg.retries,

package/dist/llm/inference/index.js

@@ -12,6 +12,7 @@
  */
 import "./providers/index.js";
 import { getInferenceProvider, listInferenceProviders, registerInferenceProvider, } from "./registry.js";
+import { firstNonEmptyString } from "../util.js";
 import { estimateTokens, writeTelemetry } from "../telemetry.js";
 const DEFAULT_TIMEOUT_MS = 30_000;
 const DEFAULT_RETRIES = 2;
@@ -27,7 +28,7 @@ export function initLLM(opts) {
     resolved = {
         provider: provider.name,
         model: opts.config.model ?? provider.defaults.model,
-        baseUrl: (opts.config.baseUrl ?? provider.defaults.baseUrl ?? "").replace(/\/+$/, ""),
+        baseUrl: (firstNonEmptyString(opts.config.baseUrl, provider.defaults.baseUrl) ?? "").replace(/\/+$/, ""),
         apiKey: opts.config.apiKey ?? null,
         fallback: opts.config.fallback ?? null,
         extra: opts.config.extra ?? {},

package/dist/llm/util.d.ts

@@ -0,0 +1,12 @@
+/**
+ * Tiny helpers shared by the embedding and inference layers.
+ */
+/**
+ * Returns the first argument that is a non-empty trimmed string, or undefined.
+ *
+ * Useful when a value may come from a config layer that normalises absent
+ * fields to `""` instead of leaving them undefined — `??` alone won't fall
+ * through `""`, but this will. See issue #131.
+ */
+export declare function firstNonEmptyString(...candidates: Array<string | undefined | null>): string | undefined;
+//# sourceMappingURL=util.d.ts.map

package/dist/llm/util.js

@@ -0,0 +1,18 @@
+/**
+ * Tiny helpers shared by the embedding and inference layers.
+ */
+/**
+ * Returns the first argument that is a non-empty trimmed string, or undefined.
+ *
+ * Useful when a value may come from a config layer that normalises absent
+ * fields to `""` instead of leaving them undefined — `??` alone won't fall
+ * through `""`, but this will. See issue #131.
+ */
+export function firstNonEmptyString(...candidates) {
+    for (const c of candidates) {
+        if (typeof c === "string" && c.trim().length > 0)
+            return c;
+    }
+    return undefined;
+}
+//# sourceMappingURL=util.js.map