graphile-llm 0.7.3 → 0.9.0

package/config-cache.js

@@ -0,0 +1,148 @@
+"use strict";
+/**
+ * config-cache — Per-database LLM billing configuration cache
+ *
+ * Caches resolved billing function names per database_id.
+ * Uses an LRU cache with TTL so config changes propagate within a bounded window
+ * without requiring a server restart.
+ *
+ * Resolution flow:
+ *   Billing config from `metaschema_modules_public.billing_module`
+ *   (schema name + function names for record_usage, check_billing_quota)
+ *
+ * All queries run through the Graphile `withPgClient` callback, which gives us
+ * a client connected to the tenant database with proper role settings.
+ *
+ * The LLM module config (provider, model, etc.) is already resolved by the
+ * LlmModulePlugin at schema-build time. This cache handles the runtime-only
+ * billing piece.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.getLlmBillingConfig = getLlmBillingConfig;
+exports.invalidateLlmBillingConfig = invalidateLlmBillingConfig;
+exports.getLlmBillingCacheStats = getLlmBillingCacheStats;
+const graphile_cache_1 = require("graphile-cache");
+// ─── SQL Queries ────────────────────────────────────────────────────────────
+/**
+ * Check if the billing_module table exists before querying it.
+ * This prevents hard errors on databases that don't have the billing
+ * module provisioned (the metaschema_modules_public schema or the
+ * billing_module table might not exist at all).
+ */
+const BILLING_MODULE_SQL = `
+  SELECT
+    s.schema_name AS public_schema,
+    ps.schema_name AS private_schema,
+    bm.record_usage_function
+  FROM metaschema_modules_public.billing_module bm
+  JOIN metaschema_public.schema s ON bm.schema_id = s.id
+  JOIN metaschema_public.schema ps ON bm.private_schema_id = ps.id
+  WHERE bm.database_id = $1
+  LIMIT 1
+`;
+/**
+ * Resolve the inference log module's schema and table name.
+ */
+const INFERENCE_LOG_MODULE_SQL = `
+  SELECT
+    s.schema_name AS schema,
+    ilm.inference_log_table_name AS table_name
+  FROM metaschema_modules_public.inference_log_module ilm
+  JOIN metaschema_public.schema s ON ilm.schema_id = s.id
+  WHERE ilm.database_id = $1
+  LIMIT 1
+`;
+// ─── Cache ──────────────────────────────────────────────────────────────────
+const billingCache = new graphile_cache_1.ModuleConfigCache({
+    name: 'billing-config',
+    ttlMs: 5 * 60 * 1000, // 5 minutes
+    max: 50
+});
+// ─── Resolution Functions ───────────────────────────────────────────────────
+/**
+ * SQL to check if a schema exists. Used as a guard before querying
+ * metaschema tables that may not be provisioned.
+ */
+const SCHEMA_EXISTS_SQL = `
+  SELECT 1 FROM information_schema.schemata WHERE schema_name = $1 LIMIT 1
+`;
+async function resolveInferenceLogConfig(pgClient, databaseId) {
+    try {
+        const schemaCheck = await pgClient.query(SCHEMA_EXISTS_SQL, ['metaschema_modules_public']);
+        if (schemaCheck.rows.length === 0)
+            return null;
+        const result = await pgClient.query(INFERENCE_LOG_MODULE_SQL, [databaseId]);
+        const row = result.rows[0];
+        if (!row?.schema || !row?.table_name)
+            return null;
+        return {
+            schema: row.schema,
+            tableName: row.table_name
+        };
+    }
+    catch {
+        return null;
+    }
+}
+async function resolveBillingConfig(pgClient, databaseId) {
+    try {
+        // Guard: check if the metaschema_modules_public schema exists.
+        // If the database doesn't have the billing module provisioned,
+        // this schema (or the billing_module table) won't exist.
+        const schemaCheck = await pgClient.query(SCHEMA_EXISTS_SQL, ['metaschema_modules_public']);
+        if (schemaCheck.rows.length === 0)
+            return null;
+        const result = await pgClient.query(BILLING_MODULE_SQL, [databaseId]);
+        const row = result.rows[0];
+        if (!row?.record_usage_function)
+            return null;
+        return {
+            publicSchema: row.public_schema,
+            privateSchema: row.private_schema,
+            recordUsageFunction: row.record_usage_function,
+            // The check_billing_quota function name follows the inflection pattern
+            checkBillingQuotaFunction: 'check_billing_quota'
+        };
+    }
+    catch {
+        // Schema/table doesn't exist or query failed — billing not available
+        return null;
+    }
+}
+// ─── Public API ─────────────────────────────────────────────────────────────
+/**
+ * Resolve billing config for a database.
+ * Results are cached per database_id with a 5-minute TTL.
+ *
+ * @param pgClient - A client connected to the tenant database (from withPgClient)
+ * @param databaseId - The database UUID
+ */
+async function getLlmBillingConfig(pgClient, databaseId) {
+    const cached = billingCache.get(databaseId);
+    if (cached)
+        return cached;
+    const [billing, inferenceLog] = await Promise.all([
+        resolveBillingConfig(pgClient, databaseId),
+        resolveInferenceLogConfig(pgClient, databaseId)
+    ]);
+    const entry = { billing, inferenceLog };
+    billingCache.set(databaseId, entry);
+    return entry;
+}
+/**
+ * Invalidate the cached config for a specific database (or all).
+ */
+function invalidateLlmBillingConfig(databaseId) {
+    if (databaseId) {
+        billingCache.delete(databaseId);
+    }
+    else {
+        billingCache.clear();
+    }
+}
+/**
+ * Get cache stats for diagnostics.
+ */
+function getLlmBillingCacheStats() {
+    return { size: billingCache.size, max: 50 };
+}

package/embedder.d.ts

@@ -24,12 +24,12 @@ export declare function buildEmbedder(config: EmbedderConfig): EmbedderFunction
  */
 export declare function buildEmbedderFromModule(data: LlmModuleData): EmbedderFunction | null;
 /**
- * Resolve an embedder from environment variables via getEnvOptions().
+ * Resolve an embedder from environment variables.
  * This is a fallback for development when no llm_module or defaultEmbedder is configured.
  *
- * Environment variables (parsed by @constructive-io/graphql-env):
- *   EMBEDDER_PROVIDER - Provider name ('ollama')
- *   EMBEDDER_MODEL    - Model identifier
- *   EMBEDDER_BASE_URL - Provider base URL
+ * Environment variables (with defaults from env.ts):
+ *   EMBEDDER_PROVIDER  - Provider name (default: 'ollama')
+ *   EMBEDDER_MODEL     - Model identifier (default: 'nomic-embed-text')
+ *   EMBEDDER_BASE_URL  - Provider base URL (default: 'http://localhost:11434')
  */
 export declare function buildEmbedderFromEnv(): EmbedderFunction | null;

package/embedder.js

@@ -18,10 +18,12 @@ exports.buildEmbedder = buildEmbedder;
 exports.buildEmbedderFromModule = buildEmbedderFromModule;
 exports.buildEmbedderFromEnv = buildEmbedderFromEnv;
 const ollama_1 = __importDefault(require("@agentic-kit/ollama"));
-const graphql_env_1 = require("@constructive-io/graphql-env");
+const env_1 = require("./env");
 // ─── Built-in Providers ─────────────────────────────────────────────────────
 /**
  * Create an Ollama-based embedder function.
+ *
+ * Uses the /api/embed endpoint which returns prompt_eval_count (real token count).
  */
 function createOllamaEmbedder(baseUrl = 'http://localhost:11434', model = 'nomic-embed-text') {
     const client = new ollama_1.default(baseUrl);
@@ -55,27 +57,19 @@ function buildEmbedderFromModule(data) {
     return buildEmbedder({
         provider: data.embedding_provider,
         model: data.embedding_model,
-        baseUrl: data.embedding_base_url,
-        apiKey: data.api_key_ref,
+        baseUrl: data.embedding_base_url
     });
 }
 /**
- * Resolve an embedder from environment variables via getEnvOptions().
+ * Resolve an embedder from environment variables.
  * This is a fallback for development when no llm_module or defaultEmbedder is configured.
  *
- * Environment variables (parsed by @constructive-io/graphql-env):
- *   EMBEDDER_PROVIDER - Provider name ('ollama')
- *   EMBEDDER_MODEL    - Model identifier
- *   EMBEDDER_BASE_URL - Provider base URL
+ * Environment variables (with defaults from env.ts):
+ *   EMBEDDER_PROVIDER  - Provider name (default: 'ollama')
+ *   EMBEDDER_MODEL     - Model identifier (default: 'nomic-embed-text')
+ *   EMBEDDER_BASE_URL  - Provider base URL (default: 'http://localhost:11434')
  */
 function buildEmbedderFromEnv() {
-    const { llm } = (0, graphql_env_1.getEnvOptions)();
-    const provider = llm?.embedder?.provider;
-    if (!provider)
-        return null;
-    return buildEmbedder({
-        provider,
-        model: llm?.embedder?.model,
-        baseUrl: llm?.embedder?.baseUrl,
-    });
+    const { embedding } = (0, env_1.getLlmEnvOptions)();
+    return buildEmbedder(embedding);
 }

package/env.d.ts

@@ -0,0 +1,31 @@
+/**
+ * LLM Environment Configuration
+ *
+ * Single source of truth for all LLM-related environment variables and defaults.
+ * Every other module in graphile-llm imports from here — no direct process.env
+ * reads or scattered null coalescing elsewhere.
+ *
+ * Environment variables:
+ *   EMBEDDER_PROVIDER  - Embedding provider name ('ollama')
+ *   EMBEDDER_MODEL     - Embedding model (default: 'nomic-embed-text')
+ *   EMBEDDER_BASE_URL  - Embedding provider URL (default: 'http://localhost:11434')
+ *   CHAT_PROVIDER      - Chat provider name ('ollama')
+ *   CHAT_MODEL         - Chat model (default: 'llama3')
+ *   CHAT_BASE_URL      - Chat provider URL (default: 'http://localhost:11434')
+ */
+export interface LlmProviderConfig {
+    provider: string;
+    model: string;
+    baseUrl: string;
+}
+export interface LlmEnvOptions {
+    embedding: LlmProviderConfig;
+    chat: LlmProviderConfig;
+}
+/**
+ * Resolve LLM configuration from environment variables with sensible defaults.
+ *
+ * Call this once and pass the result around — never read process.env directly
+ * in plugin code.
+ */
+export declare function getLlmEnvOptions(): LlmEnvOptions;

package/env.js

@@ -0,0 +1,52 @@
+"use strict";
+/**
+ * LLM Environment Configuration
+ *
+ * Single source of truth for all LLM-related environment variables and defaults.
+ * Every other module in graphile-llm imports from here — no direct process.env
+ * reads or scattered null coalescing elsewhere.
+ *
+ * Environment variables:
+ *   EMBEDDER_PROVIDER  - Embedding provider name ('ollama')
+ *   EMBEDDER_MODEL     - Embedding model (default: 'nomic-embed-text')
+ *   EMBEDDER_BASE_URL  - Embedding provider URL (default: 'http://localhost:11434')
+ *   CHAT_PROVIDER      - Chat provider name ('ollama')
+ *   CHAT_MODEL         - Chat model (default: 'llama3')
+ *   CHAT_BASE_URL      - Chat provider URL (default: 'http://localhost:11434')
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.getLlmEnvOptions = getLlmEnvOptions;
+// ─── Defaults ───────────────────────────────────────────────────────────────
+const LLM_DEFAULTS = {
+    embedding: {
+        provider: 'ollama',
+        model: 'nomic-embed-text',
+        baseUrl: 'http://localhost:11434'
+    },
+    chat: {
+        provider: 'ollama',
+        model: 'llama3',
+        baseUrl: 'http://localhost:11434'
+    }
+};
+// ─── Resolution ─────────────────────────────────────────────────────────────
+/**
+ * Resolve LLM configuration from environment variables with sensible defaults.
+ *
+ * Call this once and pass the result around — never read process.env directly
+ * in plugin code.
+ */
+function getLlmEnvOptions() {
+    return {
+        embedding: {
+            provider: process.env.EMBEDDER_PROVIDER ?? LLM_DEFAULTS.embedding.provider,
+            model: process.env.EMBEDDER_MODEL ?? LLM_DEFAULTS.embedding.model,
+            baseUrl: process.env.EMBEDDER_BASE_URL ?? LLM_DEFAULTS.embedding.baseUrl
+        },
+        chat: {
+            provider: process.env.CHAT_PROVIDER ?? LLM_DEFAULTS.chat.provider,
+            model: process.env.CHAT_MODEL ?? LLM_DEFAULTS.chat.model,
+            baseUrl: process.env.CHAT_BASE_URL ?? LLM_DEFAULTS.chat.baseUrl
+        }
+    };
+}