@oneuptime/common 9.2.16 → 9.2.18

package/Server/Utils/AI/IncidentAIContextBuilder.ts

@@ -0,0 +1,710 @@
+import ObjectID from "../../../Types/ObjectID";
+import Incident from "../../../Models/DatabaseModels/Incident";
+import IncidentStateTimeline from "../../../Models/DatabaseModels/IncidentStateTimeline";
+import IncidentInternalNote from "../../../Models/DatabaseModels/IncidentInternalNote";
+import IncidentPublicNote from "../../../Models/DatabaseModels/IncidentPublicNote";
+import IncidentService from "../../Services/IncidentService";
+import IncidentStateTimelineService from "../../Services/IncidentStateTimelineService";
+import IncidentInternalNoteService from "../../Services/IncidentInternalNoteService";
+import IncidentPublicNoteService from "../../Services/IncidentPublicNoteService";
+import WorkspaceUtil, { WorkspaceChannelMessage } from "../Workspace/Workspace";
+import WorkspaceProjectAuthTokenService from "../../Services/WorkspaceProjectAuthTokenService";
+import WorkspaceProjectAuthToken from "../../../Models/DatabaseModels/WorkspaceProjectAuthToken";
+import logger from "../Logger";
+import CaptureSpan from "../Telemetry/CaptureSpan";
+import OneUptimeDate from "../../../Types/Date";
+import SortOrder from "../../../Types/BaseDatabase/SortOrder";
+import { LLMMessage } from "../LLM/LLMService";
+import NotificationRuleWorkspaceChannel from "../../../Types/Workspace/NotificationRules/NotificationRuleWorkspaceChannel";
+import WorkspaceType from "../../../Types/Workspace/WorkspaceType";
+
+export interface IncidentContextData {
+  incident: Incident;
+  stateTimeline: Array<IncidentStateTimeline>;
+  internalNotes: Array<IncidentInternalNote>;
+  publicNotes: Array<IncidentPublicNote>;
+  workspaceMessages: Array<WorkspaceChannelMessage>;
+}
+
+export interface AIGenerationContext {
+  contextText: string;
+  systemPrompt: string;
+  messages: Array<LLMMessage>;
+}
+
+export default class IncidentAIContextBuilder {
+  @CaptureSpan()
+  public static async buildIncidentContext(data: {
+    incidentId: ObjectID;
+    includeWorkspaceMessages?: boolean;
+    workspaceMessageLimit?: number;
+  }): Promise<IncidentContextData> {
+    const incident: Incident | null = await IncidentService.findOneById({
+      id: data.incidentId,
+      select: {
+        _id: true,
+        title: true,
+        description: true,
+        createdAt: true,
+        postmortemNote: true,
+        remediationNotes: true,
+        rootCause: true,
+        customFields: true,
+        projectId: true,
+        incidentSeverity: {
+          name: true,
+          color: true,
+        },
+        currentIncidentState: {
+          name: true,
+          color: true,
+        },
+        monitors: {
+          name: true,
+        },
+        labels: {
+          name: true,
+          color: true,
+        },
+        postUpdatesToWorkspaceChannels: true,
+      },
+      props: {
+        isRoot: true,
+      },
+    });
+
+    if (!incident) {
+      throw new Error("Incident not found");
+    }
+
+    // Fetch state timeline
+    const stateTimeline: Array<IncidentStateTimeline> =
+      await IncidentStateTimelineService.findBy({
+        query: {
+          incidentId: data.incidentId,
+        },
+        select: {
+          _id: true,
+          createdAt: true,
+          startsAt: true,
+          endsAt: true,
+          rootCause: true,
+          incidentState: {
+            name: true,
+            color: true,
+          },
+          createdByUser: {
+            name: true,
+            email: true,
+          },
+        },
+        sort: {
+          startsAt: SortOrder.Ascending,
+        },
+        limit: 100,
+        skip: 0,
+        props: {
+          isRoot: true,
+        },
+      });
+
+    // Fetch internal notes
+    const internalNotes: Array<IncidentInternalNote> =
+      await IncidentInternalNoteService.findBy({
+        query: {
+          incidentId: data.incidentId,
+        },
+        select: {
+          _id: true,
+          note: true,
+          createdAt: true,
+          createdByUser: {
+            name: true,
+            email: true,
+          },
+        },
+        sort: {
+          createdAt: SortOrder.Ascending,
+        },
+        limit: 100,
+        skip: 0,
+        props: {
+          isRoot: true,
+        },
+      });
+
+    // Fetch public notes
+    const publicNotes: Array<IncidentPublicNote> =
+      await IncidentPublicNoteService.findBy({
+        query: {
+          incidentId: data.incidentId,
+        },
+        select: {
+          _id: true,
+          note: true,
+          createdAt: true,
+          postedAt: true,
+          createdByUser: {
+            name: true,
+            email: true,
+          },
+        },
+        sort: {
+          createdAt: SortOrder.Ascending,
+        },
+        limit: 100,
+        skip: 0,
+        props: {
+          isRoot: true,
+        },
+      });
+
+    // Fetch workspace messages if requested and channels exist
+    let workspaceMessages: Array<WorkspaceChannelMessage> = [];
+
+    const workspaceChannels:
+      | Array<NotificationRuleWorkspaceChannel>
+      | undefined = incident.postUpdatesToWorkspaceChannels as
+      | Array<NotificationRuleWorkspaceChannel>
+      | undefined;
+
+    if (
+      data.includeWorkspaceMessages &&
+      workspaceChannels &&
+      workspaceChannels.length > 0 &&
+      incident.projectId
+    ) {
+      try {
+        const fetchParams: {
+          projectId: ObjectID;
+          workspaceChannels: Array<NotificationRuleWorkspaceChannel>;
+          limit?: number;
+          oldestTimestamp?: Date;
+        } = {
+          projectId: incident.projectId,
+          workspaceChannels: workspaceChannels,
+          limit: data.workspaceMessageLimit || 500,
+        };
+
+        if (incident.createdAt) {
+          fetchParams.oldestTimestamp = incident.createdAt;
+        }
+
+        workspaceMessages =
+          await this.getWorkspaceMessagesForIncident(fetchParams);
+      } catch (error) {
+        logger.error(`Error fetching workspace messages: ${error}`);
+        // Continue without workspace messages
+      }
+    }
+
+    return {
+      incident,
+      stateTimeline,
+      internalNotes,
+      publicNotes,
+      workspaceMessages,
+    };
+  }
+
+  @CaptureSpan()
+  public static formatIncidentContextForPostmortem(
+    contextData: IncidentContextData,
+    template?: string,
+  ): AIGenerationContext {
+    const {
+      incident,
+      stateTimeline,
+      internalNotes,
+      publicNotes,
+      workspaceMessages,
+    } = contextData;
+
+    let contextText: string = "";
+
+    // Basic incident information
+    contextText += "# Incident Information\n\n";
+    contextText += `**Title:** ${incident.title || "N/A"}\n\n`;
+    contextText += `**Description:** ${incident.description || "N/A"}\n\n`;
+    contextText += `**Severity:** ${incident.incidentSeverity?.name || "N/A"}\n\n`;
+    contextText += `**Current State:** ${incident.currentIncidentState?.name || "N/A"}\n\n`;
+    contextText += `**Created At:** ${incident.createdAt ? OneUptimeDate.getDateAsFormattedString(incident.createdAt) : "N/A"}\n\n`;
+
+    // Affected monitors
+    if (incident.monitors && incident.monitors.length > 0) {
+      contextText += "**Affected Monitors:** ";
+      contextText += incident.monitors
+        .map((m: { name?: string }) => {
+          return m.name;
+        })
+        .join(", ");
+      contextText += "\n\n";
+    }
+
+    // Labels
+    if (incident.labels && incident.labels.length > 0) {
+      contextText += "**Labels:** ";
+      contextText += incident.labels
+        .map((l: { name?: string }) => {
+          return l.name;
+        })
+        .join(", ");
+      contextText += "\n\n";
+    }
+
+    // Root cause if available
+    if (incident.rootCause) {
+      contextText += `**Root Cause:** ${incident.rootCause}\n\n`;
+    }
+
+    // Remediation notes if available
+    if (incident.remediationNotes) {
+      contextText += `**Remediation Notes:** ${incident.remediationNotes}\n\n`;
+    }
+
+    // State timeline
+    if (stateTimeline.length > 0) {
+      contextText += "# State Timeline\n\n";
+      for (const timeline of stateTimeline) {
+        const startTime: string = timeline.startsAt
+          ? OneUptimeDate.getDateAsFormattedString(timeline.startsAt)
+          : "N/A";
+        const stateName: string =
+          timeline.incidentState?.name?.toString() || "Unknown";
+        const createdBy: string =
+          timeline.createdByUser?.name?.toString() ||
+          timeline.createdByUser?.email?.toString() ||
+          "System";
+
+        contextText += `- **${startTime}**: State changed to **${stateName}** by ${createdBy}\n`;
+        if (timeline.rootCause) {
+          contextText += `  - Root cause noted: ${timeline.rootCause}\n`;
+        }
+      }
+      contextText += "\n";
+    }
+
+    // Internal notes
+    if (internalNotes.length > 0) {
+      contextText += "# Internal Notes (Private)\n\n";
+      for (const note of internalNotes) {
+        const noteTime: string = note.createdAt
+          ? OneUptimeDate.getDateAsFormattedString(note.createdAt)
+          : "N/A";
+        const createdBy: string =
+          note.createdByUser?.name?.toString() ||
+          note.createdByUser?.email?.toString() ||
+          "Unknown";
+
+        contextText += `**[${noteTime}] ${createdBy}:**\n`;
+        contextText += `${note.note || "N/A"}\n\n`;
+      }
+    }
+
+    // Public notes
+    if (publicNotes.length > 0) {
+      contextText += "# Public Notes\n\n";
+      for (const note of publicNotes) {
+        const noteTime: string = note.postedAt
+          ? OneUptimeDate.getDateAsFormattedString(note.postedAt)
+          : note.createdAt
+            ? OneUptimeDate.getDateAsFormattedString(note.createdAt)
+            : "N/A";
+        const createdBy: string =
+          note.createdByUser?.name?.toString() ||
+          note.createdByUser?.email?.toString() ||
+          "Unknown";
+
+        contextText += `**[${noteTime}] ${createdBy}:**\n`;
+        contextText += `${note.note || "N/A"}\n\n`;
+      }
+    }
+
+    // Workspace messages (Slack/Teams)
+    if (workspaceMessages.length > 0) {
+      contextText += "# Discussion from Incident Channel\n\n";
+      contextText += WorkspaceUtil.formatMessagesAsContext(workspaceMessages, {
+        includeTimestamp: true,
+        includeUsername: true,
+        maxLength: 30000,
+      });
+      contextText += "\n\n";
+    }
+
+    // System prompt for postmortem generation
+    let systemPrompt: string;
+
+    if (template) {
+      // When a template is provided, strictly fill only the template
+      systemPrompt = `You are an expert Site Reliability Engineer (SRE) and incident response specialist. Your task is to fill in an incident postmortem template based on the provided incident data.
+
+CRITICAL INSTRUCTIONS:
+- You MUST use ONLY the exact template structure provided below
+- Fill in each section of the template with relevant information from the incident data
+- Do NOT add any new sections, headers, or content that is not part of the template
+- Do NOT add introductions, conclusions, or any text outside the template structure
+- If a section in the template has no relevant data, write "No data available" or leave the placeholder text
+- Be blameless - focus on systemic improvements rather than individual blame
+- Write in a professional, clear, and concise manner
+
+TEMPLATE TO FILL (use this exact structure):
+
+${template}`;
+    } else {
+      // When no template is provided, use standard format
+      systemPrompt = `You are an expert Site Reliability Engineer (SRE) and incident response specialist. Your task is to generate a comprehensive, well-structured incident postmortem based on the provided incident data.
+
+The postmortem should:
+1. Be written in a blameless manner, focusing on systemic improvements rather than individual blame
+2. Include a clear executive summary
+3. Provide a detailed timeline of events
+4. Identify the root cause(s) and contributing factors
+5. Outline the impact on users and systems
+6. List actionable items to prevent recurrence
+7. Include lessons learned
+
+Use a standard incident postmortem format with sections for: Executive Summary, Timeline, Root Cause Analysis, Impact, Action Items, and Lessons Learned.
+
+Write in a professional, clear, and concise manner. Use markdown formatting for better readability.`;
+    }
+
+    // Build user message based on whether template is provided
+    const userMessage: string = template
+      ? `Fill in the template above using ONLY the following incident data. Output only the filled template, nothing else:\n\n${contextText}`
+      : `Based on the following incident data, please generate a comprehensive incident postmortem:\n\n${contextText}`;
+
+    // Build messages array
+    const messages: Array<LLMMessage> = [
+      {
+        role: "system",
+        content: systemPrompt,
+      },
+      {
+        role: "user",
+        content: userMessage,
+      },
+    ];
+
+    return {
+      contextText,
+      systemPrompt,
+      messages,
+    };
+  }
+
+  @CaptureSpan()
+  public static formatIncidentContextForNote(
+    contextData: IncidentContextData,
+    noteType: "public" | "internal",
+    template?: string,
+  ): AIGenerationContext {
+    const {
+      incident,
+      stateTimeline,
+      internalNotes,
+      publicNotes,
+      workspaceMessages,
+    } = contextData;
+
+    let contextText: string = "";
+
+    // Basic incident information
+    contextText += "# Incident Information\n\n";
+    contextText += `**Title:** ${incident.title || "N/A"}\n\n`;
+    contextText += `**Description:** ${incident.description || "N/A"}\n\n`;
+    contextText += `**Severity:** ${incident.incidentSeverity?.name || "N/A"}\n\n`;
+    contextText += `**Current State:** ${incident.currentIncidentState?.name || "N/A"}\n\n`;
+    contextText += `**Created At:** ${incident.createdAt ? OneUptimeDate.getDateAsFormattedString(incident.createdAt) : "N/A"}\n\n`;
+
+    // Affected monitors
+    if (incident.monitors && incident.monitors.length > 0) {
+      contextText += "**Affected Monitors:** ";
+      contextText += incident.monitors
+        .map((m: { name?: string }) => {
+          return m.name;
+        })
+        .join(", ");
+      contextText += "\n\n";
+    }
+
+    // Labels
+    if (incident.labels && incident.labels.length > 0) {
+      contextText += "**Labels:** ";
+      contextText += incident.labels
+        .map((l: { name?: string }) => {
+          return l.name;
+        })
+        .join(", ");
+      contextText += "\n\n";
+    }
+
+    // Root cause if available
+    if (incident.rootCause) {
+      contextText += `**Root Cause:** ${incident.rootCause}\n\n`;
+    }
+
+    // State timeline
+    if (stateTimeline.length > 0) {
+      contextText += "# State Timeline\n\n";
+      for (const timeline of stateTimeline) {
+        const startTime: string = timeline.startsAt
+          ? OneUptimeDate.getDateAsFormattedString(timeline.startsAt)
+          : "N/A";
+        const stateName: string =
+          timeline.incidentState?.name?.toString() || "Unknown";
+        const createdBy: string =
+          timeline.createdByUser?.name?.toString() ||
+          timeline.createdByUser?.email?.toString() ||
+          "System";
+
+        contextText += `- **${startTime}**: State changed to **${stateName}** by ${createdBy}\n`;
+        if (timeline.rootCause) {
+          contextText += `  - Root cause noted: ${timeline.rootCause}\n`;
+        }
+      }
+      contextText += "\n";
+    }
+
+    // Include internal notes for context (for both note types)
+    if (internalNotes.length > 0) {
+      contextText += "# Internal Notes (Private)\n\n";
+      for (const note of internalNotes) {
+        const noteTime: string = note.createdAt
+          ? OneUptimeDate.getDateAsFormattedString(note.createdAt)
+          : "N/A";
+        const createdBy: string =
+          note.createdByUser?.name?.toString() ||
+          note.createdByUser?.email?.toString() ||
+          "Unknown";
+
+        contextText += `**[${noteTime}] ${createdBy}:**\n`;
+        contextText += `${note.note || "N/A"}\n\n`;
+      }
+    }
+
+    // Public notes
+    if (publicNotes.length > 0) {
+      contextText += "# Public Notes\n\n";
+      for (const note of publicNotes) {
+        const noteTime: string = note.postedAt
+          ? OneUptimeDate.getDateAsFormattedString(note.postedAt)
+          : note.createdAt
+            ? OneUptimeDate.getDateAsFormattedString(note.createdAt)
+            : "N/A";
+        const createdBy: string =
+          note.createdByUser?.name?.toString() ||
+          note.createdByUser?.email?.toString() ||
+          "Unknown";
+
+        contextText += `**[${noteTime}] ${createdBy}:**\n`;
+        contextText += `${note.note || "N/A"}\n\n`;
+      }
+    }
+
+    // Workspace messages (Slack/Teams)
+    if (workspaceMessages.length > 0) {
+      contextText += "# Discussion from Incident Channel\n\n";
+      contextText += WorkspaceUtil.formatMessagesAsContext(workspaceMessages, {
+        includeTimestamp: true,
+        includeUsername: true,
+        maxLength: 20000,
+      });
+      contextText += "\n\n";
+    }
+
+    // System prompt for note generation
+    let systemPrompt: string;
+
+    if (noteType === "public") {
+      if (template) {
+        systemPrompt = `You are an expert incident communicator. Your task is to fill in a public incident note template based on the provided incident data.
+
+CRITICAL INSTRUCTIONS:
+- You MUST use ONLY the exact template structure provided below
+- Fill in each section of the template with relevant information from the incident data
+- Do NOT add any new sections, headers, or content that is not part of the template
+- Do NOT add introductions, conclusions, or any text outside the template structure
+- Write in a professional, clear, and customer-friendly manner
+- Focus on what customers need to know: impact, current status, and next steps
+- Avoid technical jargon - keep it understandable for non-technical readers
+- Be concise but informative
+
+TEMPLATE TO FILL (use this exact structure):
+
+${template}`;
+      } else {
+        systemPrompt = `You are an expert incident communicator. Your task is to generate a public incident note that will be visible to customers on the status page.
+
+The note should:
+1. Be written in a professional, customer-friendly tone
+2. Clearly communicate the current status of the incident
+3. Explain the impact on users without excessive technical details
+4. Provide information about what is being done to resolve the issue
+5. Set appropriate expectations about resolution timing if known
+6. Be concise but informative
+
+DO NOT include:
+- Internal technical details that customers don't need
+- Blame or finger-pointing
+- Confidential information
+- Excessive jargon
+
+Write in markdown format for better readability.`;
+      }
+    } else if (template) {
+      // Internal note with template
+      systemPrompt = `You are an expert Site Reliability Engineer (SRE). Your task is to fill in an internal incident note template based on the provided incident data.
+
+CRITICAL INSTRUCTIONS:
+- You MUST use ONLY the exact template structure provided below
+- Fill in each section of the template with relevant information from the incident data
+- Do NOT add any new sections, headers, or content that is not part of the template
+- Do NOT add introductions, conclusions, or any text outside the template structure
+- Be technical and detailed - this is for the internal team
+- Include relevant technical details, metrics, and observations
+
+TEMPLATE TO FILL (use this exact structure):
+
+${template}`;
+    } else {
+      // Internal note without template
+      systemPrompt = `You are an expert Site Reliability Engineer (SRE). Your task is to generate an internal incident note for the response team.
+
+The note should:
+1. Provide technical details about the current situation
+2. Document observations, findings, or actions taken
+3. Include relevant metrics or error messages if mentioned in the context
+4. Be detailed enough to help team members understand the situation
+5. Use technical language appropriate for the engineering team
+
+Write in markdown format for better readability. Be thorough and technical.`;
+    }
+
+    // Build user message
+    const userMessage: string = template
+      ? `Fill in the template above using ONLY the following incident data. Output only the filled template, nothing else:\n\n${contextText}`
+      : `Based on the following incident data, please generate ${noteType === "public" ? "a customer-facing public" : "an internal technical"} incident note:\n\n${contextText}`;
+
+    // Build messages array
+    const messages: Array<LLMMessage> = [
+      {
+        role: "system",
+        content: systemPrompt,
+      },
+      {
+        role: "user",
+        content: userMessage,
+      },
+    ];
+
+    return {
+      contextText,
+      systemPrompt,
+      messages,
+    };
+  }
+
+  @CaptureSpan()
+  public static buildGenericAIContext(data: {
+    systemPrompt: string;
+    userPrompt: string;
+    context?: string;
+  }): AIGenerationContext {
+    let userContent: string = data.userPrompt;
+
+    if (data.context) {
+      userContent = `${data.userPrompt}\n\nContext:\n${data.context}`;
+    }
+
+    const messages: Array<LLMMessage> = [
+      {
+        role: "system",
+        content: data.systemPrompt,
+      },
+      {
+        role: "user",
+        content: userContent,
+      },
+    ];
+
+    return {
+      contextText: data.context || "",
+      systemPrompt: data.systemPrompt,
+      messages,
+    };
+  }
+
+  @CaptureSpan()
+  private static async getWorkspaceMessagesForIncident(data: {
+    projectId: ObjectID;
+    workspaceChannels: Array<NotificationRuleWorkspaceChannel>;
+    limit?: number;
+    oldestTimestamp?: Date;
+  }): Promise<Array<WorkspaceChannelMessage>> {
+    const allMessages: Array<WorkspaceChannelMessage> = [];
+
+    for (const channel of data.workspaceChannels) {
+      try {
+        // Get auth token for this workspace type
+        const projectAuth: WorkspaceProjectAuthToken | null =
+          await WorkspaceProjectAuthTokenService.getProjectAuth({
+            projectId: data.projectId,
+            workspaceType: channel.workspaceType,
+          });
+
+        if (!projectAuth || !projectAuth.authToken) {
+          logger.debug(
+            `No auth token found for workspace type: ${channel.workspaceType}`,
+          );
+          continue;
+        }
+
+        const messagesParams: {
+          channelId: string;
+          authToken: string;
+          projectId: ObjectID;
+          workspaceType: WorkspaceType;
+          teamId?: string;
+          limit?: number;
+          oldestTimestamp?: Date;
+        } = {
+          channelId: channel.id,
+          authToken: projectAuth.authToken,
+          projectId: data.projectId,
+          workspaceType: channel.workspaceType,
+        };
+
+        if (channel.teamId) {
+          messagesParams.teamId = channel.teamId;
+        }
+
+        if (data.limit !== undefined) {
+          messagesParams.limit = data.limit;
+        }
+
+        if (data.oldestTimestamp) {
+          messagesParams.oldestTimestamp = data.oldestTimestamp;
+        }
+
+        const messages: Array<WorkspaceChannelMessage> =
+          await WorkspaceUtil.getChannelMessages(messagesParams);
+
+        allMessages.push(...messages);
+      } catch (error) {
+        logger.error(
+          `Error fetching messages from channel ${channel.id}: ${error}`,
+        );
+        // Continue with other channels even if one fails
+      }
+    }
+
+    // Sort all messages by timestamp
+    allMessages.sort(
+      (a: WorkspaceChannelMessage, b: WorkspaceChannelMessage) => {
+        return a.timestamp.getTime() - b.timestamp.getTime();
+      },
+    );
+
+    return allMessages;
+  }
+}