sanity checking, do not duplicate responses

prisma/migrations/20250705032242_add_processing_column/migration.sql

@@ -0,0 +1,18 @@
+-- RedefineTables
+PRAGMA defer_foreign_keys=ON;
+PRAGMA foreign_keys=OFF;
+CREATE TABLE "new_Response" (
+    "id" INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT,
+    "pleromaNotificationId" TEXT NOT NULL DEFAULT 'null',
+    "to" TEXT NOT NULL,
+    "request" TEXT,
+    "response" TEXT,
+    "createdAt" DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    "processedAt" DATETIME,
+    "isProcessing" BOOLEAN NOT NULL DEFAULT true
+);
+INSERT INTO "new_Response" ("createdAt", "id", "pleromaNotificationId", "processedAt", "request", "response", "to") SELECT "createdAt", "id", "pleromaNotificationId", "processedAt", "request", "response", "to" FROM "Response";
+DROP TABLE "Response";
+ALTER TABLE "new_Response" RENAME TO "Response";
+PRAGMA foreign_keys=ON;
+PRAGMA defer_foreign_keys=OFF;

prisma/migrations/20250705035205_add_default_values_to_db/migration.sql

@@ -0,0 +1,18 @@
+-- RedefineTables
+PRAGMA defer_foreign_keys=ON;
+PRAGMA foreign_keys=OFF;
+CREATE TABLE "new_Response" (
+    "id" INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT,
+    "pleromaNotificationId" TEXT NOT NULL DEFAULT 'null',
+    "to" TEXT NOT NULL DEFAULT 'null',
+    "request" TEXT NOT NULL DEFAULT 'null',
+    "response" TEXT NOT NULL DEFAULT 'null',
+    "createdAt" DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    "processedAt" DATETIME,
+    "isProcessing" BOOLEAN NOT NULL DEFAULT true
+);
+INSERT INTO "new_Response" ("createdAt", "id", "isProcessing", "pleromaNotificationId", "processedAt", "request", "response", "to") SELECT "createdAt", "id", "isProcessing", "pleromaNotificationId", "processedAt", coalesce("request", 'null') AS "request", coalesce("response", 'null') AS "response", "to" FROM "Response";
+DROP TABLE "Response";
+ALTER TABLE "new_Response" RENAME TO "Response";
+PRAGMA foreign_keys=ON;
+PRAGMA defer_foreign_keys=OFF;

prisma/schema.prisma

@@ -14,11 +14,12 @@ datasource db {
 model Response {
   id                    Int       @id @default(autoincrement())
   pleromaNotificationId String    @default("null")
-  to                    String
+  to                    String    @default("null")
-  request               String?
+  request               String    @default("null")
-  response              String?
+  response              String    @default("null")
   createdAt             DateTime  @default(now())
   processedAt           DateTime?
+  isProcessing          Boolean   @default(true)
 }
 
 model User {

src/main.ts

@@ -3,6 +3,7 @@ import {
   OllamaResponse,
   NewStatusBody,
   Notification,
+  OllamaConfigOptions,
 } from "../types.js";
 import striptags from "striptags";
 import { PrismaClient } from "../generated/prisma/client.js";
@@ -51,7 +52,7 @@ const alreadyRespondedTo = async (
 ): Promise<boolean> => {
   try {
     const duplicate = await prisma.response.findFirst({
-      where: { pleromaNotificationId: notification.status.id },
+      where: { pleromaNotificationId: notification.id, isProcessing: true },
     });
     if (duplicate) {
       return true;
@@ -67,12 +68,13 @@ const storePromptData = async (
   ollamaResponseBody: OllamaResponse
 ) => {
   try {
-    await prisma.response.create({
+    await prisma.response.updateMany({
+      where: { pleromaNotificationId: notification.id },
       data: {
         response: ollamaResponseBody.response,
-        request: striptags(notification.status.content),
+        request: trimInputData(notification.status.content),
         to: notification.account.fqn,
-        pleromaNotificationId: notification.status.id,
+        isProcessing: false,
       },
     });
   } catch (error: any) {
@@ -87,6 +89,18 @@ const trimInputData = (input: string) => {
   return split.slice(promptStringIndex + 1).join(" "); // returns everything after the !prompt
 };
 
+const recordPendingResponse = async (notification: Notification) => {
+  try {
+    await prisma.response.create({
+      data: {
+        pleromaNotificationId: notification.id,
+      },
+    });
+  } catch (error: any) {
+    throw new Error(error.message);
+  }
+};
+
 const generateOllamaRequest = async (
   notification: Notification
 ): Promise<OllamaResponse | undefined> => {
@@ -107,7 +121,12 @@ const generateOllamaRequest = async (
       if (await alreadyRespondedTo(notification)) {
         return;
       }
+      await recordPendingResponse(notification);
       await storeUserData(notification);
+      const ollamaConfig: OllamaConfigOptions = {
+        temperature: 1.2,
+        num_predict: 400,
+      };
       const ollamaRequestBody: OllamaRequest = {
         model: process.env.OLLAMA_MODEL as string,
         system: process.env.OLLAMA_SYSTEM_PROMPT as string,
@@ -115,6 +134,7 @@ const generateOllamaRequest = async (
           notification.status.content
         )}`,
         stream: false,
+        options: ollamaConfig,
       };
       const response = await fetch(`${process.env.OLLAMA_URL}/api/generate`, {
         method: "POST",
@@ -177,6 +197,11 @@ const deleteNotification = async (notification: Notification) => {
     if (!notification.id) {
       return;
     }
+    await prisma.response.updateMany({
+      // this is probably not the best way to do this, but since we may have duplicate notifications, we have to update all of them - probably won't scale (lmao)
+      where: { pleromaNotificationId: notification.id },
+      data: { isProcessing: false },
+    });
     const response = await fetch(
       `${process.env.PLEROMA_INSTANCE_URL}/api/v1/notifications/${notification.id}/dismiss`,
       {

types.d.ts

@@ -69,15 +69,28 @@ export interface Mention {
   username: string;
 }
 
+/**
+ * Experimental settings, I wouldn't recommend messing with these if you don't know how they work (I don't either)
+ */
 export interface OllamaConfigOptions {
+  /**
+   * Number of tokens guaranteed to be kept in memory during response generation. Higher values leave less
+   * possible room for num_ctx
+   */
   num_keep?: number;
   seed?: number;
+  /**
+   * Sets maximum of tokens in the response
+   */
   num_predict?: number;
   top_k?: number;
   top_p?: number;
   min_p?: number;
   typical_p?: number;
   repeat_last_n?: number;
+  /**
+   * How close of a response should the response be to the original prompt - lower = more focused response
+   */
   temperature?: number;
   repeat_penalty?: number;
   presence_penalty?: number;
@@ -88,6 +101,9 @@ export interface OllamaConfigOptions {
   penalize_newline?: boolean;
   stop?: string[];
   numa?: boolean;
+  /**
+   * Number of tokens for the prompt to keep in memory for the response, minus the value of num_keep
+   */
   num_ctx?: number;
   num_batch?: number;
   num_gpu?: number;