graphile-llm 0.11.5 → 0.12.0

package/esm/plugins/llm-module-plugin.js

@@ -11,10 +11,15 @@
  * to consume.
  *
  * Resolution order for the embedder:
- *   1. `llm_module` from api_modules (per-database, loaded at schema build time)
- *   2. `defaultEmbedder` from preset options (dev/testing fallback)
- *   3. Environment variables (EMBEDDER_PROVIDER, EMBEDDER_MODEL, EMBEDDER_BASE_URL)
- *   4. null — LLM features are disabled
+ *   1. `defaultEmbedder` from preset options (build-time)
+ *   2. Environment variables (EMBEDDER_PROVIDER, EMBEDDER_MODEL, EMBEDDER_BASE_URL)
+ *   3. null — LLM features are disabled
+ *
+ * Per-database model/baseUrl overrides (from `llm_module` via `ctx.useLlm()`)
+ * are applied at request time via `llmConfigStore` (AsyncLocalStorage) in
+ * the text-search-plugin resolver wrapper. The same embedder function — and
+ * its metering wrapper — handles every request; only the model name and
+ * base URL parameters change per-tenant.
  *
  * This plugin is intentionally pure — no billing or metering logic.
  * The optional LlmMeteringPlugin wraps the embedder with billing integration

package/esm/plugins/rag-plugin.js

@@ -70,18 +70,18 @@ function parseHasChunksTag(raw, codec) {
  */
 function discoverChunkTables(build) {
     const chunkTables = [];
-    const pgRegistry = build.pgRegistry;
+    const pgRegistry = build.input?.pgRegistry ?? build.pgRegistry;
     if (!pgRegistry)
         return chunkTables;
     // Scan all codecs for @hasChunks smart tag
-    for (const source of Object.values(pgRegistry.pgResources || {})) {
-        const codec = source?.codec;
-        if (!codec?.attributes)
+    for (const codec of Object.values(pgRegistry.pgCodecs || {})) {
+        const c = codec;
+        if (!c?.attributes)
             continue;
-        const tags = codec.extensions?.tags;
+        const tags = c.extensions?.tags;
         if (!tags?.hasChunks)
             continue;
-        const info = parseHasChunksTag(tags.hasChunks, codec);
+        const info = parseHasChunksTag(tags.hasChunks, c);
         if (info) {
             chunkTables.push(info);
         }
@@ -135,18 +135,6 @@ export function createLlmRagPlugin(ragDefaults = {}) {
     let embedder = null;
     let chatCompleter = null;
     const schemaExtension = extendSchema((build) => {
-        // Discover chunk-aware tables from pgRegistry
-        chunkTables = discoverChunkTables(build);
-        embedder = build.llmEmbedder || null;
-        chatCompleter = build.llmChatCompleter || null;
-        if (chunkTables.length > 0) {
-            console.log(`[graphile-llm] RAG plugin discovered ${chunkTables.length} chunk-aware table(s): ` +
-                chunkTables.map((t) => t.parentCodecName).join(', '));
-        }
-        else {
-            console.log('[graphile-llm] RAG plugin found no @hasChunks tables. ' +
-                'ragQuery will still work if chunks tables are queried directly.');
-        }
         return {
             typeDefs: gql `
         """A source chunk retrieved during RAG context assembly."""
@@ -325,8 +313,8 @@ export function createLlmRagPlugin(ragDefaults = {}) {
                 }
             }
         };
-    });
-    return {
+    }, 'LlmRagPlugin');
+    const plugin = {
         ...schemaExtension,
         name: 'LlmRagPlugin',
         version: '0.1.0',
@@ -338,4 +326,26 @@ export function createLlmRagPlugin(ragDefaults = {}) {
             'VectorCodecPlugin'
         ]
     };
+    // Wrap the build hook to also discover chunk tables.
+    // The build hook runs after all init hooks (including smart tag injection),
+    // so @hasChunks tags are guaranteed to be visible.
+    const existingBuildHook = plugin.schema.hooks.build;
+    plugin.schema.hooks.build = (build) => {
+        // Run extendSchema's build hook first (sets up graphql ref)
+        build = existingBuildHook(build) || build;
+        // Discover chunk tables — runs during build phase when all smart tags are applied
+        chunkTables = discoverChunkTables(build);
+        embedder = build.llmEmbedder || null;
+        chatCompleter = build.llmChatCompleter || null;
+        if (chunkTables.length > 0) {
+            console.log(`[graphile-llm] RAG plugin discovered ${chunkTables.length} chunk-aware table(s): ` +
+                chunkTables.map((t) => t.parentCodecName).join(', '));
+        }
+        else {
+            console.log('[graphile-llm] RAG plugin found no @hasChunks tables. ' +
+                'ragQuery will still work if chunks tables are queried directly.');
+        }
+        return build;
+    };
+    return plugin;
 }

package/esm/plugins/text-mutation-plugin.js

@@ -83,8 +83,21 @@ export function createLlmTextMutationPlugin() {
                  */
                 GraphQLInputObjectType_fields(fields, build, context) {
                     const { scope: { isPgPatch, isPgBaseInput, isMutationInput, pgCodec } } = context;
-                    // Only intercept create/update input types for table rows
-                    if (!pgCodec?.attributes || (!isPgPatch && !isPgBaseInput && !isMutationInput)) {
+                    // Only intercept mutation-related input types for table rows.
+                    // PostGraphile v5 sets isPgPatch on update types, isPgBaseInput on
+                    // create base types. For tables with `serial PRIMARY KEY`, the create
+                    // input type (e.g. ArticleInput) has pgCodec set but none of the
+                    // explicit mutation scope flags — detect via type name convention.
+                    if (!pgCodec?.attributes)
+                        return fields;
+                    const typeName = context.Self.name;
+                    const hasExplicitMutationScope = isPgPatch || isPgBaseInput || isMutationInput;
+                    const looksLikeMutationInput = typeName.endsWith('Input') || typeName.endsWith('Patch');
+                    const isFilterOrCondition = typeName.endsWith('FilterInput')
+                        || typeName.endsWith('Filter')
+                        || typeName.endsWith('Condition')
+                        || typeName === 'VectorNearbyInput';
+                    if (!hasExplicitMutationScope && (!looksLikeMutationInput || isFilterOrCondition)) {
                         return fields;
                     }
                     const { graphql: { GraphQLString } } = build;

package/esm/plugins/text-search-plugin.d.ts

@@ -24,8 +24,10 @@
  * (so the schema is stable) but will return a clear error at execution time.
  *
  * If the embedder returns null (e.g. quota exceeded when the metering
- * plugin is loaded), the text field is silently removed — the query
- * continues with text-only search as a graceful fallback.
+ * plugin is loaded), behavior depends on `onQuotaExceeded`:
+ * - `'degrade'` (default): silently removes the text field and continues
+ *   with text-only search as a graceful fallback.
+ * - `'throw'`: always throws an error, even if text adapters could handle it.
  */
 import type { GraphileConfig } from 'graphile-config';
 declare global {
@@ -35,6 +37,14 @@ declare global {
         }
     }
 }
+/**
+ * Recursively walk a `where` argument object and embed any VectorNearbyInput
+ * values that have `text` instead of `vector`.
+ *
+ * If the embedder returns null (e.g. quota exceeded), the text field is
+ * removed so the pgvector filter is skipped — graceful text-only fallback.
+ */
+export declare function embedTextInWhere(obj: any, embedder: (text: string) => Promise<number[] | null>, hasTextAdapters: boolean, onQuotaExceeded?: 'degrade' | 'throw'): Promise<void>;
 /**
  * Creates the LlmTextSearchPlugin.
  *
@@ -42,4 +52,6 @@ declare global {
  * existing `vector` field. When a user provides `text`, the plugin's
  * resolver wrapper embeds it before passing to pgvector.
  */
-export declare function createLlmTextSearchPlugin(): GraphileConfig.Plugin;
+export declare function createLlmTextSearchPlugin(options?: {
+    onQuotaExceeded?: 'degrade' | 'throw';
+}): GraphileConfig.Plugin;

package/esm/plugins/text-search-plugin.js

@@ -24,9 +24,12 @@
  * (so the schema is stable) but will return a clear error at execution time.
  *
  * If the embedder returns null (e.g. quota exceeded when the metering
- * plugin is loaded), the text field is silently removed — the query
- * continues with text-only search as a graceful fallback.
+ * plugin is loaded), behavior depends on `onQuotaExceeded`:
+ * - `'degrade'` (default): silently removes the text field and continues
+ *   with text-only search as a graceful fallback.
+ * - `'throw'`: always throws an error, even if text adapters could handle it.
  */
+import { llmConfigStore } from '../embedder';
 /**
  * Check if a codec has any pgvector `vector` columns.
  */
@@ -46,12 +49,34 @@ function hasVectorColumns(pgCodec) {
  * If the embedder returns null (e.g. quota exceeded), the text field is
  * removed so the pgvector filter is skipped — graceful text-only fallback.
  */
-async function embedTextInWhere(obj, embedder) {
+export async function embedTextInWhere(obj, embedder, hasTextAdapters, onQuotaExceeded = 'degrade') {
     if (!obj || typeof obj !== 'object')
         return;
     const pending = [];
     for (const key of Object.keys(obj)) {
         const value = obj[key];
+        // Handle unifiedSearch: embed text and transform to { __text, __vector }
+        if (key === 'unifiedSearch' && typeof value === 'string' && value.trim().length > 0) {
+            pending.push((async () => {
+                const startTime = Date.now();
+                const vector = await embedder(value);
+                const latencyMs = Date.now() - startTime;
+                if (vector === null) {
+                    if (onQuotaExceeded === 'throw' || !hasTextAdapters) {
+                        throw new Error('unifiedSearch: embedding failed (quota exceeded or provider unavailable). ' +
+                            (!hasTextAdapters
+                                ? 'No text search adapters available for fallback. '
+                                : 'onQuotaExceeded is set to \'throw\'. ') +
+                            'Upgrade your plan or adjust onQuotaExceeded to \'degrade\' for text-only fallback.');
+                    }
+                    return;
+                }
+                console.log(`[graphile-llm] unifiedSearch embed: dims=${vector.length}, latency=${latencyMs}ms`);
+                // Transform to object shape that graphile-search understands
+                obj[key] = { __text: value, __vector: vector };
+            })());
+            continue;
+        }
         if (!value || typeof value !== 'object')
             continue;
         // Detect VectorNearbyInput shape: has `text` and no `vector`
@@ -61,7 +86,10 @@ async function embedTextInWhere(obj, embedder) {
                 const vector = await embedder(value.text);
                 const latencyMs = Date.now() - startTime;
                 if (vector === null) {
-                    // Embedder returned null (e.g. quota exceeded) — skip vector search
+                    if (onQuotaExceeded === 'throw') {
+                        throw new Error('VectorNearbyInput: embedding failed (quota exceeded or provider unavailable). ' +
+                            'Upgrade your plan or adjust onQuotaExceeded to \'degrade\' for graceful fallback.');
+                    }
                     delete value.text;
                     return;
                 }
@@ -74,12 +102,12 @@ async function embedTextInWhere(obj, embedder) {
         }
         // Recurse into nested filter objects (AND, OR, etc.)
         if (!Array.isArray(value)) {
-            pending.push(embedTextInWhere(value, embedder));
+            pending.push(embedTextInWhere(value, embedder, hasTextAdapters, onQuotaExceeded));
         }
         else {
             // Handle arrays (e.g. AND: [...], OR: [...])
             for (const item of value) {
-                pending.push(embedTextInWhere(item, embedder));
+                pending.push(embedTextInWhere(item, embedder, hasTextAdapters, onQuotaExceeded));
             }
         }
     }
@@ -87,6 +115,32 @@ async function embedTextInWhere(obj, embedder) {
         await Promise.all(pending);
     }
 }
+/**
+ * Resolve per-DB LLM config overrides from express-context.
+ * Returns an LlmConfigOverrides object if per-DB config is available,
+ * or null to use defaults. These overrides are string parameters only
+ * (model name, base URL) — the embedder function and its metering
+ * wrapper stay the same.
+ */
+async function resolveConfigOverrides(graphqlContext) {
+    const ctx = graphqlContext?.constructive;
+    if (!ctx?.useLlm)
+        return null;
+    try {
+        const llm = await ctx.useLlm();
+        if (!llm?.embeddingModel)
+            return null;
+        return {
+            embeddingModel: llm.embeddingModel,
+            embeddingBaseUrl: llm.embeddingBaseUrl,
+            chatModel: llm.chatModel ?? undefined,
+            chatBaseUrl: llm.chatBaseUrl ?? undefined,
+        };
+    }
+    catch {
+        return null;
+    }
+}
 /**
  * Creates the LlmTextSearchPlugin.
  *
@@ -94,7 +148,8 @@ async function embedTextInWhere(obj, embedder) {
  * existing `vector` field. When a user provides `text`, the plugin's
  * resolver wrapper embeds it before passing to pgvector.
  */
-export function createLlmTextSearchPlugin() {
+export function createLlmTextSearchPlugin(options = {}) {
+    const { onQuotaExceeded = 'degrade' } = options;
     return {
         name: 'LlmTextSearchPlugin',
         version: '0.2.0',
@@ -113,8 +168,8 @@ export function createLlmTextSearchPlugin() {
                  * The field is optional — clients provide either `text` or `vector`.
                  */
                 GraphQLInputObjectType_fields(fields, build, context) {
-                    const { scope: { inputObjectTypeName } } = context;
-                    if (inputObjectTypeName !== 'VectorNearbyInput') {
+                    const typeName = context.Self.name;
+                    if (typeName !== 'VectorNearbyInput') {
                         return fields;
                     }
                     const { graphql: { GraphQLString } } = build;
@@ -129,35 +184,51 @@ export function createLlmTextSearchPlugin() {
                 },
                 /**
                  * Wrap connection query resolvers to intercept `where` arguments that
-                 * contain VectorNearbyInput with `text`, embed the text, and replace
-                 * it with the resulting vector before the plan executes.
+                 * contain VectorNearbyInput with `text` or `unifiedSearch` with text,
+                 * embed the text, and inject the resulting vector before the plan executes.
+                 *
+                 * For tables with vector columns: embeds VectorNearbyInput.text → vector
+                 * For ALL tables with unifiedSearch: embeds unifiedSearch text → { __text, __vector }
                  *
                  * Uses the same v4-style resolver wrapping pattern as graphile-upload-plugin
                  * and graphile-bucket-provisioner-plugin.
                  */
                 GraphQLObjectType_fields_field(field, build, context) {
                     const { scope: { isRootQuery, pgCodec } } = context;
-                    // Only wrap root query fields on tables with vector columns
-                    if (!isRootQuery || !pgCodec || !hasVectorColumns(pgCodec)) {
+                    if (!isRootQuery || !pgCodec)
+                        return field;
+                    // Wrap if the table has vector columns OR has any searchable columns
+                    // (for unifiedSearch embedding support)
+                    const hasVector = hasVectorColumns(pgCodec);
+                    const hasSearchableColumns = pgCodec.attributes && Object.values(pgCodec.attributes).some((attr) => attr.codec?.name === 'tsvector' || attr.codec?.name === 'vector');
+                    if (!hasVector && !hasSearchableColumns)
                         return field;
-                    }
                     const embedder = build.llmEmbedder;
                     if (!embedder)
                         return field;
+                    // Determine if this table has text-based search adapters for fallback logic
+                    const hasTextAdapters = pgCodec.attributes && Object.values(pgCodec.attributes).some((attr) => attr.codec?.name === 'tsvector');
                     const defaultResolver = (obj) => obj[context.scope.fieldName];
                     const { resolve: oldResolve = defaultResolver, ...rest } = field;
                     return {
                         ...rest,
                         async resolve(source, args, graphqlContext, info) {
-                            // If the query has a `where` argument, check for text fields
-                            if (args?.where) {
-                                await embedTextInWhere(args.where, embedder);
-                            }
-                            // Also handle `filter` for relay-style connections
-                            if (args?.filter) {
-                                await embedTextInWhere(args.filter, embedder);
-                            }
-                            return oldResolve(source, args, graphqlContext, info);
+                            // Resolve per-DB model/baseUrl overrides from express-context.
+                            // These are just string parameters — the embedder function
+                            // (and its metering wrapper) stays the same.
+                            const overrides = await resolveConfigOverrides(graphqlContext);
+                            // Run within llmConfigStore so the embedder picks up per-DB
+                            // model/baseUrl at call time. Metering wraps the same function
+                            // and is unaffected.
+                            return llmConfigStore.run(overrides, async () => {
+                                if (args?.where) {
+                                    await embedTextInWhere(args.where, embedder, !!hasTextAdapters, onQuotaExceeded);
+                                }
+                                if (args?.filter) {
+                                    await embedTextInWhere(args.filter, embedder, !!hasTextAdapters, onQuotaExceeded);
+                                }
+                                return oldResolve(source, args, graphqlContext, info);
+                            });
                         }
                     };
                 },

package/esm/preset.d.ts

@@ -10,8 +10,8 @@
  * - Adds `{column}Text: String` companion fields on mutation inputs for vector columns
  * - Optionally enables billing/metering via the LlmMeteringPlugin
  *
- * This preset is standalone — it is NOT included in ConstructivePreset by default.
- * Projects that want LLM features opt in by adding it to their preset.
+ * Included in ConstructivePreset when `enableLlm` is true (the default).
+ * Can also be used standalone by adding it directly to your preset's `extends` array.
  *
  * @example
  * ```typescript
@@ -53,6 +53,17 @@
  * })
  * ```
  *
+ * @example With strict quota enforcement (fail if embedding unavailable):
+ * ```typescript
+ * GraphileLlmPreset({
+ *   defaultEmbedder: { provider: 'openai', model: 'text-embedding-3-small' },
+ *   metering: true,
+ *   onQuotaExceeded: 'throw',
+ *   // → if billing quota is exceeded, queries with unifiedSearch throw an error
+ *   // → without this, queries silently fall back to text-only search
+ * })
+ * ```
+ *
  * @example With custom entity_id resolution (bill per-database):
  * ```typescript
  * GraphileLlmPreset({

package/esm/preset.js

@@ -10,8 +10,8 @@
  * - Adds `{column}Text: String` companion fields on mutation inputs for vector columns
  * - Optionally enables billing/metering via the LlmMeteringPlugin
  *
- * This preset is standalone — it is NOT included in ConstructivePreset by default.
- * Projects that want LLM features opt in by adding it to their preset.
+ * Included in ConstructivePreset when `enableLlm` is true (the default).
+ * Can also be used standalone by adding it directly to your preset's `extends` array.
  *
  * @example
  * ```typescript
@@ -53,6 +53,17 @@
  * })
  * ```
  *
+ * @example With strict quota enforcement (fail if embedding unavailable):
+ * ```typescript
+ * GraphileLlmPreset({
+ *   defaultEmbedder: { provider: 'openai', model: 'text-embedding-3-small' },
+ *   metering: true,
+ *   onQuotaExceeded: 'throw',
+ *   // → if billing quota is exceeded, queries with unifiedSearch throw an error
+ *   // → without this, queries silently fall back to text-only search
+ * })
+ * ```
+ *
  * @example With custom entity_id resolution (bill per-database):
  * ```typescript
  * GraphileLlmPreset({
@@ -75,7 +86,7 @@ import { createLlmTextSearchPlugin } from './plugins/text-search-plugin';
  * @returns A GraphileConfig.Preset to add to your extends array
  */
 export function GraphileLlmPreset(options = {}) {
-    const { enableTextSearch = true, enableTextMutations = true, enableRag = false, ragDefaults, metering } = options;
+    const { enableTextSearch = true, enableTextMutations = true, enableRag = false, onQuotaExceeded, ragDefaults, metering } = options;
     const plugins = [
         createLlmModulePlugin(options)
     ];
@@ -86,7 +97,7 @@ export function GraphileLlmPreset(options = {}) {
         plugins.push(createLlmMeteringPlugin(meteringConfig));
     }
     if (enableTextSearch) {
-        plugins.push(createLlmTextSearchPlugin());
+        plugins.push(createLlmTextSearchPlugin({ onQuotaExceeded }));
     }
     if (enableTextMutations) {
         plugins.push(createLlmTextMutationPlugin());

package/esm/types.d.ts

@@ -216,6 +216,19 @@ export interface GraphileLlmOptions {
      * @default true
      */
     enableTextSearch?: boolean;
+    /**
+     * Controls behavior when the embedder returns null (e.g. billing quota exceeded).
+     *
+     * - `'degrade'` (default): Skip the vector search silently and fall back to
+     *   text-only adapters (tsvector, BM25, trgm). If no text adapters are
+     *   available, throws an error regardless of this setting.
+     * - `'throw'`: Always throw an error when the embedder fails, even if
+     *   text adapters could handle the query. Use this when semantic search
+     *   is critical and keyword-only results are unacceptable.
+     *
+     * @default 'degrade'
+     */
+    onQuotaExceeded?: 'degrade' | 'throw';
     /**
      * Whether to add `*Text` companion fields on mutation inputs for vector columns.
      * @default true

package/index.d.ts

@@ -29,17 +29,18 @@
  * };
  * ```
  */
-export type { LlmEnvOptions, LlmProviderConfig } from './env';
-export { getLlmEnvOptions } from './env';
+export type { LlmEnvOptions, LlmProviderConfig, ResolvedLlmEnvOptions } from './env';
+export { getEnvVars, getEnvOptions, getLlmEnvOptions, llmDefaults } from './env';
 export { GraphileLlmPreset } from './preset';
 export { createLlmModulePlugin } from './plugins/llm-module-plugin';
 export { createLlmRagPlugin } from './plugins/rag-plugin';
 export { createLlmTextMutationPlugin } from './plugins/text-mutation-plugin';
-export { createLlmTextSearchPlugin } from './plugins/text-search-plugin';
+export { createLlmTextSearchPlugin, embedTextInWhere } from './plugins/text-search-plugin';
 export { createLlmMeteringPlugin } from './plugins/metering-plugin';
 export type { AgentDiscovery, AgentTableInfo } from './plugins/agent-discovery-plugin';
 export { clearAgentDiscoveryCache, getAgentDiscovery } from './plugins/agent-discovery-plugin';
-export { buildEmbedder, buildEmbedderFromEnv, buildEmbedderFromModule } from './embedder';
+export type { LlmConfigOverrides } from './embedder';
+export { buildEmbedder, buildEmbedderFromEnv, buildEmbedderFromModule, llmConfigStore } from './embedder';
 export { buildChatCompleter, buildChatCompleterFromEnv, buildChatCompleterFromModule } from './chat';
 export type { InferenceLogEntry, MeteringContext, MeteringOptions, MeterResult, WithPgClient } from './metering';
 export { logInferenceUsage, meteredChat, meteredEmbed, QuotaExceededError } from './metering';

package/index.js

@@ -31,9 +31,12 @@
  * ```
  */
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.invalidateLlmBillingConfig = exports.getLlmBillingConfig = exports.getLlmBillingCacheStats = exports.QuotaExceededError = exports.meteredEmbed = exports.meteredChat = exports.logInferenceUsage = exports.buildChatCompleterFromModule = exports.buildChatCompleterFromEnv = exports.buildChatCompleter = exports.buildEmbedderFromModule = exports.buildEmbedderFromEnv = exports.buildEmbedder = exports.getAgentDiscovery = exports.clearAgentDiscoveryCache = exports.createLlmMeteringPlugin = exports.createLlmTextSearchPlugin = exports.createLlmTextMutationPlugin = exports.createLlmRagPlugin = exports.createLlmModulePlugin = exports.GraphileLlmPreset = exports.getLlmEnvOptions = void 0;
+exports.invalidateLlmBillingConfig = exports.getLlmBillingConfig = exports.getLlmBillingCacheStats = exports.QuotaExceededError = exports.meteredEmbed = exports.meteredChat = exports.logInferenceUsage = exports.buildChatCompleterFromModule = exports.buildChatCompleterFromEnv = exports.buildChatCompleter = exports.llmConfigStore = exports.buildEmbedderFromModule = exports.buildEmbedderFromEnv = exports.buildEmbedder = exports.getAgentDiscovery = exports.clearAgentDiscoveryCache = exports.createLlmMeteringPlugin = exports.embedTextInWhere = exports.createLlmTextSearchPlugin = exports.createLlmTextMutationPlugin = exports.createLlmRagPlugin = exports.createLlmModulePlugin = exports.GraphileLlmPreset = exports.llmDefaults = exports.getLlmEnvOptions = exports.getEnvOptions = exports.getEnvVars = void 0;
 var env_1 = require("./env");
+Object.defineProperty(exports, "getEnvVars", { enumerable: true, get: function () { return env_1.getEnvVars; } });
+Object.defineProperty(exports, "getEnvOptions", { enumerable: true, get: function () { return env_1.getEnvOptions; } });
 Object.defineProperty(exports, "getLlmEnvOptions", { enumerable: true, get: function () { return env_1.getLlmEnvOptions; } });
+Object.defineProperty(exports, "llmDefaults", { enumerable: true, get: function () { return env_1.llmDefaults; } });
 // Preset (recommended entry point)
 var preset_1 = require("./preset");
 Object.defineProperty(exports, "GraphileLlmPreset", { enumerable: true, get: function () { return preset_1.GraphileLlmPreset; } });
@@ -46,17 +49,18 @@ var text_mutation_plugin_1 = require("./plugins/text-mutation-plugin");
 Object.defineProperty(exports, "createLlmTextMutationPlugin", { enumerable: true, get: function () { return text_mutation_plugin_1.createLlmTextMutationPlugin; } });
 var text_search_plugin_1 = require("./plugins/text-search-plugin");
 Object.defineProperty(exports, "createLlmTextSearchPlugin", { enumerable: true, get: function () { return text_search_plugin_1.createLlmTextSearchPlugin; } });
+Object.defineProperty(exports, "embedTextInWhere", { enumerable: true, get: function () { return text_search_plugin_1.embedTextInWhere; } });
 // Metering plugin (opt-in billing integration)
 var metering_plugin_1 = require("./plugins/metering-plugin");
 Object.defineProperty(exports, "createLlmMeteringPlugin", { enumerable: true, get: function () { return metering_plugin_1.createLlmMeteringPlugin; } });
 var agent_discovery_plugin_1 = require("./plugins/agent-discovery-plugin");
 Object.defineProperty(exports, "clearAgentDiscoveryCache", { enumerable: true, get: function () { return agent_discovery_plugin_1.clearAgentDiscoveryCache; } });
 Object.defineProperty(exports, "getAgentDiscovery", { enumerable: true, get: function () { return agent_discovery_plugin_1.getAgentDiscovery; } });
-// Embedder utilities
 var embedder_1 = require("./embedder");
 Object.defineProperty(exports, "buildEmbedder", { enumerable: true, get: function () { return embedder_1.buildEmbedder; } });
 Object.defineProperty(exports, "buildEmbedderFromEnv", { enumerable: true, get: function () { return embedder_1.buildEmbedderFromEnv; } });
 Object.defineProperty(exports, "buildEmbedderFromModule", { enumerable: true, get: function () { return embedder_1.buildEmbedderFromModule; } });
+Object.defineProperty(exports, "llmConfigStore", { enumerable: true, get: function () { return embedder_1.llmConfigStore; } });
 // Chat completion utilities
 var chat_1 = require("./chat");
 Object.defineProperty(exports, "buildChatCompleter", { enumerable: true, get: function () { return chat_1.buildChatCompleter; } });

package/metering.d.ts

@@ -90,9 +90,7 @@ export interface InferenceLogEntry {
 }
 /**
  * Write a row to the usage_log_inference table.
- * Gracefully skips if the inference_log_module is not provisioned.
- *
- * TODO: Also write to child (generated) database when dual-write is needed.
+ * Delegates to the shared BillingClient from express-context.
  */
 export declare function logInferenceUsage(ctx: MeteringContext, entry: InferenceLogEntry): Promise<void>;
 /**