graphile-llm 0.8.0 → 0.10.0

package/esm/__tests__/graphile-llm.test.js

@@ -1,16 +1,16 @@
-import { join } from 'path';
 import OllamaClient from '@agentic-kit/ollama';
-import { getConnections, seed } from 'graphile-test';
 import { ConnectionFilterPreset } from 'graphile-connection-filter';
+import { createPgvectorAdapter } from 'graphile-search/adapters/pgvector';
 import { VectorCodecPlugin } from 'graphile-search/codecs/vector-codec';
 import { createUnifiedSearchPlugin } from 'graphile-search/plugin';
-import { createPgvectorAdapter } from 'graphile-search/adapters/pgvector';
+import { getConnections, seed } from 'graphile-test';
+import { join } from 'path';
+import { buildChatCompleter, buildChatCompleterFromEnv, buildChatCompleterFromModule } from '../../src/chat';
+import { buildEmbedder, buildEmbedderFromEnv, buildEmbedderFromModule } from '../../src/embedder';
 import { createLlmModulePlugin } from '../../src/plugins/llm-module-plugin';
-import { createLlmTextSearchPlugin } from '../../src/plugins/text-search-plugin';
-import { createLlmTextMutationPlugin } from '../../src/plugins/text-mutation-plugin';
 import { createLlmRagPlugin } from '../../src/plugins/rag-plugin';
-import { buildEmbedder, buildEmbedderFromModule, buildEmbedderFromEnv, } from '../../src/embedder';
-import { buildChatCompleter, buildChatCompleterFromModule, buildChatCompleterFromEnv, } from '../../src/chat';
+import { createLlmTextMutationPlugin } from '../../src/plugins/text-mutation-plugin';
+import { createLlmTextSearchPlugin } from '../../src/plugins/text-search-plugin';
 // ─── @agentic-kit/ollama client ─────────────────────────────────────────────
 const ollamaClient = new OllamaClient('http://localhost:11434');
 async function ensureNomicModel() {
@@ -30,14 +30,14 @@ describe('Embedder abstraction', () => {
             const embedder = buildEmbedder({
                 provider: 'ollama',
                 model: 'nomic-embed-text',
-                baseUrl: 'http://localhost:11434',
+                baseUrl: 'http://localhost:11434'
             });
             expect(embedder).not.toBeNull();
             expect(typeof embedder).toBe('function');
         });
         it('returns null for unknown provider', () => {
             const embedder = buildEmbedder({
-                provider: 'unknown-provider',
+                provider: 'unknown-provider'
             });
             expect(embedder).toBeNull();
         });
@@ -52,7 +52,7 @@ describe('Embedder abstraction', () => {
             const moduleData = {
                 embedding_provider: 'ollama',
                 embedding_model: 'nomic-embed-text',
-                embedding_base_url: 'http://localhost:11434',
+                embedding_base_url: 'http://localhost:11434'
             };
             const embedder = buildEmbedderFromModule(moduleData);
             expect(embedder).not.toBeNull();
@@ -60,7 +60,7 @@ describe('Embedder abstraction', () => {
         });
         it('returns null for unsupported provider in module data', () => {
             const moduleData = {
-                embedding_provider: 'unsupported',
+                embedding_provider: 'unsupported'
             };
             const embedder = buildEmbedderFromModule(moduleData);
             expect(embedder).toBeNull();
@@ -83,7 +83,7 @@ describe('Embedder abstraction', () => {
                 ...originalEnv,
                 EMBEDDER_PROVIDER: 'ollama',
                 EMBEDDER_MODEL: 'nomic-embed-text',
-                EMBEDDER_BASE_URL: 'http://localhost:11434',
+                EMBEDDER_BASE_URL: 'http://localhost:11434'
             };
             const embedder = buildEmbedderFromEnv();
             expect(embedder).not.toBeNull();
@@ -101,7 +101,7 @@ describe('graphile-llm schema enrichment', () => {
     let query;
     beforeAll(async () => {
         const unifiedPlugin = createUnifiedSearchPlugin({
-            adapters: [createPgvectorAdapter()],
+            adapters: [createPgvectorAdapter()]
         });
         const testPreset = {
             extends: [ConnectionFilterPreset()],
@@ -114,18 +114,18 @@ describe('graphile-llm schema enrichment', () => {
                     defaultEmbedder: {
                         provider: 'ollama',
                         model: 'nomic-embed-text',
-                        baseUrl: 'http://localhost:11434',
-                    },
+                        baseUrl: 'http://localhost:11434'
+                    }
                 }),
                 createLlmTextSearchPlugin(),
-                createLlmTextMutationPlugin(),
-            ],
+                createLlmTextMutationPlugin()
+            ]
         };
         const connections = await getConnections({
             schemas: ['llm_test'],
             preset: testPreset,
             useRoot: true,
-            authRole: 'postgres',
+            authRole: 'postgres'
         }, [seed.sqlfile([join(__dirname, './setup.sql')])]);
         db = connections.db;
         teardown = connections.teardown;
@@ -254,33 +254,36 @@ describe('graphile-llm with real Ollama embedding', () => {
         const embedder = buildEmbedder({
             provider: 'ollama',
             model: 'nomic-embed-text',
-            baseUrl: 'http://localhost:11434',
+            baseUrl: 'http://localhost:11434'
         });
         expect(embedder).not.toBeNull();
-        const vector = await embedder('Machine learning is transforming AI');
+        const result = await embedder('Machine learning is transforming AI');
         // nomic-embed-text produces 768-dimensional vectors
-        expect(Array.isArray(vector)).toBe(true);
-        expect(vector.length).toBe(768);
+        expect(Array.isArray(result.embedding)).toBe(true);
+        expect(result.embedding.length).toBe(768);
+        expect(result.promptTokens).toBeGreaterThan(0);
         // All elements should be numbers
-        for (const v of vector) {
+        for (const v of result.embedding) {
             expect(typeof v).toBe('number');
             expect(Number.isFinite(v)).toBe(true);
         }
         // Vector should not be all zeros
-        const magnitude = Math.sqrt(vector.reduce((sum, v) => sum + v * v, 0));
+        const magnitude = Math.sqrt(result.embedding.reduce((sum, v) => sum + v * v, 0));
         expect(magnitude).toBeGreaterThan(0);
     });
     it('should produce different vectors for semantically different text', async () => {
         const embedder = buildEmbedder({
             provider: 'ollama',
             model: 'nomic-embed-text',
-            baseUrl: 'http://localhost:11434',
+            baseUrl: 'http://localhost:11434'
         });
         expect(embedder).not.toBeNull();
-        const [vecA, vecB] = await Promise.all([
+        const [resultA, resultB] = await Promise.all([
             embedder('Artificial intelligence and machine learning'),
-            embedder('Cooking recipes for Italian pasta dishes'),
+            embedder('Cooking recipes for Italian pasta dishes')
         ]);
+        const vecA = resultA.embedding;
+        const vecB = resultB.embedding;
         expect(vecA.length).toBe(768);
         expect(vecB.length).toBe(768);
         // Compute cosine similarity
@@ -300,13 +303,15 @@ describe('graphile-llm with real Ollama embedding', () => {
         const embedder = buildEmbedder({
             provider: 'ollama',
             model: 'nomic-embed-text',
-            baseUrl: 'http://localhost:11434',
+            baseUrl: 'http://localhost:11434'
         });
         expect(embedder).not.toBeNull();
-        const [vecA, vecB] = await Promise.all([
+        const [resultA, resultB] = await Promise.all([
             embedder('Machine learning and artificial intelligence'),
-            embedder('AI and ML are subfields of computer science'),
+            embedder('AI and ML are subfields of computer science')
         ]);
+        const vecA = resultA.embedding;
+        const vecB = resultB.embedding;
         expect(vecA.length).toBe(768);
         expect(vecB.length).toBe(768);
         // Compute cosine similarity
@@ -322,11 +327,14 @@ describe('graphile-llm with real Ollama embedding', () => {
         // Semantically similar texts should have high similarity
         expect(cosineSimilarity).toBeGreaterThan(0.5);
     });
-    it('should produce embeddings via @agentic-kit/ollama OllamaClient directly', async () => {
-        const vector = await ollamaClient.generateEmbedding('Testing the agentic-kit Ollama client directly', 'nomic-embed-text');
-        expect(Array.isArray(vector)).toBe(true);
-        expect(vector.length).toBe(768);
-        for (const v of vector) {
+    it('should produce embeddings with token count via @agentic-kit/ollama OllamaClient directly', async () => {
+        const result = await ollamaClient.generateEmbedding('Testing the agentic-kit Ollama client directly', 'nomic-embed-text');
+        expect(result).toHaveProperty('embedding');
+        expect(result).toHaveProperty('promptTokens');
+        expect(Array.isArray(result.embedding)).toBe(true);
+        expect(result.embedding.length).toBe(768);
+        expect(result.promptTokens).toBeGreaterThan(0);
+        for (const v of result.embedding) {
             expect(typeof v).toBe('number');
             expect(Number.isFinite(v)).toBe(true);
         }
@@ -341,14 +349,14 @@ describe('Chat completion abstraction', () => {
             const chat = buildChatCompleter({
                 provider: 'ollama',
                 model: 'llama3',
-                baseUrl: 'http://localhost:11434',
+                baseUrl: 'http://localhost:11434'
             });
             expect(chat).not.toBeNull();
             expect(typeof chat).toBe('function');
         });
         it('returns null for unknown provider', () => {
             const chat = buildChatCompleter({
-                provider: 'unknown-provider',
+                provider: 'unknown-provider'
             });
             expect(chat).toBeNull();
         });
@@ -364,7 +372,7 @@ describe('Chat completion abstraction', () => {
                 embedding_provider: 'ollama',
                 chat_provider: 'ollama',
                 chat_model: 'llama3',
-                chat_base_url: 'http://localhost:11434',
+                chat_base_url: 'http://localhost:11434'
             };
             const chat = buildChatCompleterFromModule(moduleData);
             expect(chat).not.toBeNull();
@@ -372,7 +380,7 @@ describe('Chat completion abstraction', () => {
         });
         it('returns null when chat_provider is not set', () => {
             const moduleData = {
-                embedding_provider: 'ollama',
+                embedding_provider: 'ollama'
             };
             const chat = buildChatCompleterFromModule(moduleData);
             expect(chat).toBeNull();
@@ -395,7 +403,7 @@ describe('Chat completion abstraction', () => {
                 ...originalEnv,
                 CHAT_PROVIDER: 'ollama',
                 CHAT_MODEL: 'llama3',
-                CHAT_BASE_URL: 'http://localhost:11434',
+                CHAT_BASE_URL: 'http://localhost:11434'
             };
             const chat = buildChatCompleterFromEnv();
             expect(chat).not.toBeNull();
@@ -434,10 +442,10 @@ function makeTestSmartTagsPlugin(tagsByTable) {
                             Object.assign(c.extensions.tags, tags);
                         }
                         return _;
-                    },
-                },
-            },
-        },
+                    }
+                }
+            }
+        }
     };
 }
 describe('RAG plugin schema enrichment', () => {
@@ -446,7 +454,7 @@ describe('RAG plugin schema enrichment', () => {
     let query;
     beforeAll(async () => {
         const unifiedPlugin = createUnifiedSearchPlugin({
-            adapters: [createPgvectorAdapter()],
+            adapters: [createPgvectorAdapter()]
         });
         const smartTagsPlugin = makeTestSmartTagsPlugin({
             articles: {
@@ -455,16 +463,22 @@ describe('RAG plugin schema enrichment', () => {
                     parentFk: 'parent_id',
                     parentPk: 'id',
                     embeddingField: 'embedding',
-                    contentField: 'content',
-                },
-            },
+                    contentField: 'content'
+                }
+            }
+        });
+        // Mock embedder that returns a fixed 3-dim vector with token count
+        const mockEmbedder = async (_text) => ({
+            embedding: [1, 0, 0],
+            promptTokens: 5
         });
-        // Mock embedder that returns a fixed 3-dim vector
-        const mockEmbedder = async (_text) => [1, 0, 0];
-        // Mock chat completer that returns a canned response
+        // Mock chat completer that returns a canned response with usage
         const mockChatCompleter = async (messages) => {
             const userMessage = messages.find((m) => m.role === 'user');
-            return `Mock answer for: ${userMessage?.content || 'unknown'}`;
+            return {
+                content: `Mock answer for: ${userMessage?.content || 'unknown'}`,
+                usage: { input: 10, output: 15, reasoning: 0, cacheRead: 0, cacheWrite: 0, totalTokens: 25 }
+            };
         };
         const testPreset = {
             extends: [ConnectionFilterPreset()],
@@ -476,13 +490,13 @@ describe('RAG plugin schema enrichment', () => {
                     defaultEmbedder: {
                         provider: 'ollama',
                         model: 'nomic-embed-text',
-                        baseUrl: 'http://localhost:11434',
-                    },
+                        baseUrl: 'http://localhost:11434'
+                    }
                 }),
                 createLlmTextSearchPlugin(),
                 createLlmTextMutationPlugin(),
-                createLlmRagPlugin(),
-            ],
+                createLlmRagPlugin()
+            ]
         };
         // Override the embedder and chat completer on the build context
         // by wrapping the LlmModulePlugin's build hook
@@ -495,20 +509,20 @@ describe('RAG plugin schema enrichment', () => {
                     build(build) {
                         return build.extend(build, {
                             llmEmbedder: mockEmbedder,
-                            llmChatCompleter: mockChatCompleter,
+                            llmChatCompleter: mockChatCompleter
                         }, 'TestOverridePlugin overriding embedder and chat completer');
-                    },
-                },
-            },
+                    }
+                }
+            }
         };
         const connections = await getConnections({
             schemas: ['llm_test'],
             preset: {
                 ...testPreset,
-                plugins: [...testPreset.plugins, overridePlugin],
+                plugins: [...testPreset.plugins, overridePlugin]
             },
             useRoot: true,
-            authRole: 'postgres',
+            authRole: 'postgres'
         }, [seed.sqlfile([join(__dirname, './setup.sql')])]);
         db = connections.db;
         teardown = connections.teardown;
@@ -632,7 +646,7 @@ describe('GraphileLlmPreset toggles', () => {
     it('enableRag=false excludes RAG plugin (no ragQuery field)', async () => {
         const { GraphileLlmPreset } = await import('../../src/preset');
         const preset = GraphileLlmPreset({
-            enableRag: false,
+            enableRag: false
         });
         const pluginNames = preset.plugins.map((p) => p.name);
         expect(pluginNames).not.toContain('LlmRagPlugin');
@@ -640,7 +654,7 @@ describe('GraphileLlmPreset toggles', () => {
     it('enableRag=true includes RAG plugin', async () => {
         const { GraphileLlmPreset } = await import('../../src/preset');
         const preset = GraphileLlmPreset({
-            enableRag: true,
+            enableRag: true
         });
         const pluginNames = preset.plugins.map((p) => p.name);
         expect(pluginNames).toContain('LlmRagPlugin');
@@ -648,7 +662,7 @@ describe('GraphileLlmPreset toggles', () => {
     it('enableTextSearch=false excludes text search plugin', async () => {
         const { GraphileLlmPreset } = await import('../../src/preset');
         const preset = GraphileLlmPreset({
-            enableTextSearch: false,
+            enableTextSearch: false
         });
         const pluginNames = preset.plugins.map((p) => p.name);
         expect(pluginNames).not.toContain('LlmTextSearchPlugin');
@@ -658,7 +672,7 @@ describe('GraphileLlmPreset toggles', () => {
     it('enableTextMutations=false excludes text mutation plugin', async () => {
         const { GraphileLlmPreset } = await import('../../src/preset');
         const preset = GraphileLlmPreset({
-            enableTextMutations: false,
+            enableTextMutations: false
         });
         const pluginNames = preset.plugins.map((p) => p.name);
         expect(pluginNames).not.toContain('LlmTextMutationPlugin');
@@ -668,7 +682,7 @@ describe('GraphileLlmPreset toggles', () => {
         const preset = GraphileLlmPreset({
             enableTextSearch: false,
             enableTextMutations: false,
-            enableRag: false,
+            enableRag: false
         });
         const pluginNames = preset.plugins.map((p) => p.name);
         expect(pluginNames).toEqual(['LlmModulePlugin']);

package/esm/chat.js

@@ -11,38 +11,51 @@
  *   2. The preset's `defaultChatCompleter` option (fallback for dev/testing)
  *   3. Environment variables (CHAT_PROVIDER, CHAT_MODEL, CHAT_BASE_URL)
  */
-import OllamaClient from '@agentic-kit/ollama';
+import { OllamaAdapter } from '@agentic-kit/ollama';
 import { getLlmEnvOptions } from './env';
 // ─── Built-in Providers ─────────────────────────────────────────────────────
 /**
  * Create an Ollama-based chat completion function.
  *
- * Uses OllamaClient.generate() with a messages array, which internally
- * routes to the /api/chat endpoint.
+ * Uses OllamaAdapter.stream() to get both response content and real token
+ * usage counts from the provider (prompt_eval_count, eval_count).
  */
 function createOllamaChatCompleter(baseUrl = 'http://localhost:11434', model = 'llama3') {
-    const client = new OllamaClient(baseUrl);
+    const adapter = new OllamaAdapter(baseUrl);
     return async (messages, options) => {
-        // Build the input for OllamaClient.generate() in chat mode
-        const input = {
-            model,
-            messages: messages.filter((m) => m.role !== 'system'),
-        };
-        // Extract system message if present
         const systemMsg = messages.find((m) => m.role === 'system');
-        if (systemMsg) {
-            input.system = systemMsg.content;
-        }
-        if (options?.temperature !== undefined) {
-            input.temperature = options.temperature;
-        }
-        const startTime = Date.now();
-        const response = await client.generate(input);
-        const latencyMs = Date.now() - startTime;
-        // Token count logging (metering deferred to billing system)
-        console.log(`[graphile-llm] Chat completion: model=${model}, latency=${latencyMs}ms, ` +
-            `messages=${messages.length}`);
-        return response;
+        const nonSystem = messages.filter((m) => m.role !== 'system');
+        const modelDesc = adapter.createModel(model, {
+            maxOutputTokens: options?.maxTokens
+        });
+        const context = {
+            systemPrompt: systemMsg?.content,
+            messages: nonSystem.map((m) => ({
+                role: m.role,
+                content: m.content,
+                timestamp: Date.now()
+            }))
+        };
+        const stream = adapter.stream(modelDesc, context, {
+            temperature: options?.temperature,
+            maxTokens: options?.maxTokens
+        });
+        const result = await stream.result();
+        const content = result.content
+            .filter((block) => block.type === 'text')
+            .map((block) => block.text)
+            .join('');
+        return {
+            content,
+            usage: {
+                input: result.usage.input,
+                output: result.usage.output,
+                reasoning: result.usage.reasoning,
+                cacheRead: result.usage.cacheRead,
+                cacheWrite: result.usage.cacheWrite,
+                totalTokens: result.usage.totalTokens
+            }
+        };
     };
 }
 // ─── Chat Completer Construction ────────────────────────────────────────────
@@ -73,7 +86,7 @@ export function buildChatCompleterFromModule(data) {
     return buildChatCompleter({
         provider: data.chat_provider,
         model: data.chat_model,
-        baseUrl: data.chat_base_url,
+        baseUrl: data.chat_base_url
     });
 }
 /**

package/esm/config-cache.js

@@ -51,7 +51,7 @@ const INFERENCE_LOG_MODULE_SQL = `
 const billingCache = new ModuleConfigCache({
     name: 'billing-config',
     ttlMs: 5 * 60 * 1000, // 5 minutes
-    max: 50,
+    max: 50
 });
 // ─── Resolution Functions ───────────────────────────────────────────────────
 /**
@@ -72,7 +72,7 @@ async function resolveInferenceLogConfig(pgClient, databaseId) {
             return null;
         return {
             schema: row.schema,
-            tableName: row.table_name,
+            tableName: row.table_name
         };
     }
     catch {
@@ -96,7 +96,7 @@ async function resolveBillingConfig(pgClient, databaseId) {
             privateSchema: row.private_schema,
             recordUsageFunction: row.record_usage_function,
             // The check_billing_quota function name follows the inflection pattern
-            checkBillingQuotaFunction: 'check_billing_quota',
+            checkBillingQuotaFunction: 'check_billing_quota'
         };
     }
     catch {
@@ -118,7 +118,7 @@ export async function getLlmBillingConfig(pgClient, databaseId) {
         return cached;
     const [billing, inferenceLog] = await Promise.all([
         resolveBillingConfig(pgClient, databaseId),
-        resolveInferenceLogConfig(pgClient, databaseId),
+        resolveInferenceLogConfig(pgClient, databaseId)
     ]);
     const entry = { billing, inferenceLog };
     billingCache.set(databaseId, entry);

package/esm/embedder.js

@@ -14,6 +14,8 @@ import { getLlmEnvOptions } from './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 OllamaClient(baseUrl);
@@ -47,7 +49,7 @@ export function buildEmbedderFromModule(data) {
     return buildEmbedder({
         provider: data.embedding_provider,
         model: data.embedding_model,
-        baseUrl: data.embedding_base_url,
+        baseUrl: data.embedding_base_url
     });
 }
 /**

package/esm/env.js

@@ -18,13 +18,13 @@ const LLM_DEFAULTS = {
     embedding: {
         provider: 'ollama',
         model: 'nomic-embed-text',
-        baseUrl: 'http://localhost:11434',
+        baseUrl: 'http://localhost:11434'
     },
     chat: {
         provider: 'ollama',
         model: 'llama3',
-        baseUrl: 'http://localhost:11434',
-    },
+        baseUrl: 'http://localhost:11434'
+    }
 };
 // ─── Resolution ─────────────────────────────────────────────────────────────
 /**
@@ -38,12 +38,12 @@ export function getLlmEnvOptions() {
         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,
+            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,
-        },
+            baseUrl: process.env.CHAT_BASE_URL ?? LLM_DEFAULTS.chat.baseUrl
+        }
     };
 }

package/esm/index.d.ts

@@ -29,20 +29,20 @@
  * };
  * ```
  */
-export { getLlmEnvOptions } from './env';
 export type { LlmEnvOptions, LlmProviderConfig } from './env';
+export { getLlmEnvOptions } from './env';
 export { GraphileLlmPreset } from './preset';
 export { createLlmModulePlugin } from './plugins/llm-module-plugin';
-export { createLlmTextSearchPlugin } from './plugins/text-search-plugin';
-export { createLlmTextMutationPlugin } from './plugins/text-mutation-plugin';
 export { createLlmRagPlugin } from './plugins/rag-plugin';
+export { createLlmTextMutationPlugin } from './plugins/text-mutation-plugin';
+export { createLlmTextSearchPlugin } from './plugins/text-search-plugin';
 export { createLlmMeteringPlugin } from './plugins/metering-plugin';
-export { getAgentDiscovery, clearAgentDiscoveryCache } from './plugins/agent-discovery-plugin';
-export type { AgentTableInfo, AgentDiscovery } from './plugins/agent-discovery-plugin';
-export { buildEmbedder, buildEmbedderFromModule, buildEmbedderFromEnv, } from './embedder';
-export { buildChatCompleter, buildChatCompleterFromModule, buildChatCompleterFromEnv, } from './chat';
-export { meteredEmbed, meteredChat, logInferenceUsage, QuotaExceededError } from './metering';
-export type { MeteringContext, MeteringOptions, MeterResult, WithPgClient, InferenceLogEntry } from './metering';
-export { getLlmBillingConfig, invalidateLlmBillingConfig, getLlmBillingCacheStats, } from './config-cache';
-export type { BillingConfig, LlmBillingCacheEntry, InferenceLogConfig, PgClient } from './config-cache';
-export type { EmbedderFunction, EmbedderConfig, ChatFunction, ChatConfig, ChatMessage, ChatOptions, LlmModuleData, GraphileLlmOptions, MeteringConfig, RagDefaults, ChunkTableInfo, } from './types';
+export type { AgentDiscovery, AgentTableInfo } from './plugins/agent-discovery-plugin';
+export { clearAgentDiscoveryCache, getAgentDiscovery } from './plugins/agent-discovery-plugin';
+export { buildEmbedder, buildEmbedderFromEnv, buildEmbedderFromModule } from './embedder';
+export { buildChatCompleter, buildChatCompleterFromEnv, buildChatCompleterFromModule } from './chat';
+export type { InferenceLogEntry, MeteringContext, MeteringOptions, MeterResult, WithPgClient } from './metering';
+export { logInferenceUsage, meteredChat, meteredEmbed, QuotaExceededError } from './metering';
+export type { BillingConfig, InferenceLogConfig, LlmBillingCacheEntry, PgClient } from './config-cache';
+export { getLlmBillingCacheStats, getLlmBillingConfig, invalidateLlmBillingConfig } from './config-cache';
+export type { ChatConfig, ChatFunction, ChatMessage, ChatOptions, ChatResult, ChunkTableInfo, EmbedderConfig, EmbedderFunction, EmbeddingResult, GraphileLlmOptions, LlmModuleData, LlmUsage, MeteringConfig, RagDefaults } from './types';

package/esm/index.js

@@ -29,24 +29,20 @@
  * };
  * ```
  */
-// Environment configuration (single source of truth for LLM defaults)
 export { getLlmEnvOptions } from './env';
 // Preset (recommended entry point)
 export { GraphileLlmPreset } from './preset';
 // Individual plugins (pure — no billing dependency)
 export { createLlmModulePlugin } from './plugins/llm-module-plugin';
-export { createLlmTextSearchPlugin } from './plugins/text-search-plugin';
-export { createLlmTextMutationPlugin } from './plugins/text-mutation-plugin';
 export { createLlmRagPlugin } from './plugins/rag-plugin';
+export { createLlmTextMutationPlugin } from './plugins/text-mutation-plugin';
+export { createLlmTextSearchPlugin } from './plugins/text-search-plugin';
 // Metering plugin (opt-in billing integration)
 export { createLlmMeteringPlugin } from './plugins/metering-plugin';
-// Agent discovery (queries agent_chat_module config table at runtime)
-export { getAgentDiscovery, clearAgentDiscoveryCache } from './plugins/agent-discovery-plugin';
+export { clearAgentDiscoveryCache, getAgentDiscovery } from './plugins/agent-discovery-plugin';
 // Embedder utilities
-export { buildEmbedder, buildEmbedderFromModule, buildEmbedderFromEnv, } from './embedder';
+export { buildEmbedder, buildEmbedderFromEnv, buildEmbedderFromModule } from './embedder';
 // Chat completion utilities
-export { buildChatCompleter, buildChatCompleterFromModule, buildChatCompleterFromEnv, } from './chat';
-// Metering utilities (for custom integration)
-export { meteredEmbed, meteredChat, logInferenceUsage, QuotaExceededError } from './metering';
-// Config cache (for custom integration)
-export { getLlmBillingConfig, invalidateLlmBillingConfig, getLlmBillingCacheStats, } from './config-cache';
+export { buildChatCompleter, buildChatCompleterFromEnv, buildChatCompleterFromModule } from './chat';
+export { logInferenceUsage, meteredChat, meteredEmbed, QuotaExceededError } from './metering';
+export { getLlmBillingCacheStats, getLlmBillingConfig, invalidateLlmBillingConfig } from './config-cache';

package/esm/metering.d.ts

@@ -9,16 +9,16 @@
  * When the quota check fails, the wrapper returns null (graceful degradation)
  * instead of throwing, so the search pipeline can fall back to text-only.
  *
- * Token counts are estimated from text length (~4 chars per token). No
- * tokenizer needed — the billing system uses tokens as abstract units
- * and the credit_cost on each model's meter normalizes the relative expense.
+ * Token counts:
+ *   - Chat: real provider counts via ChatResult.usage (from OllamaAdapter.stream())
+ *   - Embedding: real provider counts via EmbeddingResult.promptTokens (from /api/embed)
  *
  * The billing functions live in the tenant database and are called via the
  * Graphile `withPgClient` callback. Function locations (schema, names) are
  * resolved from `billing_module` metaschema and cached by `config-cache.ts`.
  */
-import type { PgClient, BillingConfig, InferenceLogConfig } from './config-cache';
-import type { EmbedderFunction, ChatFunction, ChatMessage, ChatOptions } from './types';
+import type { BillingConfig, InferenceLogConfig, PgClient } from './config-cache';
+import type { ChatFunction, ChatMessage, ChatOptions, EmbedderFunction } from './types';
 /**
  * Callback matching Graphile's withPgClient signature.
  * Acquires a pg client, calls the callback, then releases the client.