Skip to content

Analyses and batches

The Analyses page (sidebar) lists every AI analysis. Each one runs on a batch: batches are the core analysis units in SerOps. Each batch contains logs, metrics, a distributed-trace summary, and AI-generated insights.

What is a Batch?

A batch is a snapshot of your system at a point in time:

Field Description
batch_id Unique identifier (e.g., batch_20260107_143000_a1b2)
timestamp When the batch was created
logs_analyzed Number of log entries
issue_count Number of issues found
environment dev, staging, or production
model_name AI model used for analysis

Batch Creation

Batches are created automatically at a configurable interval (default: every 10 minutes). Each organisation can set its own interval (5 minutes to 12 hours) under Settings → Data Management → Batch Generation Interval. See Batch Generation Interval for details.

Pipeline Overview

graph LR
    subgraph "Data Sources (external)"
        L[Log Provider<br/>ES / Loki / Datadog]
        M[Metric Provider<br/>Prometheus / VM / Mimir]
        D[Deployments<br/>via webhooks]
        C[CI/CD Logs<br/>via webhooks]
    end

    subgraph "Batch Creation Task"
        B[Aggregate &<br/>Filter Logs]
        P[Fetch & Preprocess<br/>Metrics]
        MG[Merge All<br/>Data Sources]
    end

    subgraph "AI Pipeline"
        R[PII Redaction]
        PC[Project &amp; Service<br/>Context]
        AI[AI Analysis<br/>Pluggable Models]
    end

    subgraph "Output"
        BA[AIAnalyzedBatch]
        IS[Issues]
        N[Notifications]
    end

    L -->|pull| B
    M -->|pull| P
    D --> MG
    C --> MG
    B --> MG
    P --> MG
    MG --> R --> PC --> AI
    AI --> BA
    AI --> IS --> N

What Gets Included

Data Type Source Purpose
Application Logs Log Provider Filtered by service + environment
Metrics Metric Provider Raw data for UI visualization
Metrics Summary Preprocessed AI-optimized with anomaly detection
Code Changes Deployments Commits deployed (not just committed)
CI/CD Logs Log Provider Pipeline event documents with failure traces

Batch Data Structure

The batch stores both raw data (for UI) and preprocessed data (for AI):

{
  "logs": [...],           // Filtered application logs
  "metrics": {...},        // Raw metrics arrays (for charts)
  "metrics_summary": {...}, // Preprocessed metrics (for AI)
  "code_changes": [...],   // Deployed commits with diffs
  "cicd_logs": [...],      // Pipeline logs with traces
  "statistics": {...}      // Aggregated counts
}

What AI Receives

The AI model receives a formatted text summary, not raw JSON arrays. This includes project & service context (names and descriptions) when available, giving the AI architectural awareness of your system.

Project & Service Context:
-------------
Project: "payment-platform" — E-commerce payment processing, Django + PostgreSQL
Services in scope:
  - backend-api [application]: Payment processing API, handles Stripe webhooks
  - redis-cache [infrastructure]: Session and cart caching

System Metrics Summary:
-------------
Memory: 85.7% avg, peak 90.2%, current 88%, trend: STABLE
  Total Memory: 16 GB
  Memory Anomalies: 1 high-usage periods detected
    [2026-01-21T15:30:00Z] 92.5% peak, duration: 3m 0s
      Correlation window: 15:25:00 to 15:35:00

CPU: 45.2% avg, peak 78%, current 42%, trend: STABLE
  (Based on 24 memory and 24 CPU data points)

Note: Code changes are linked via Deployments, not commit timestamps. This ensures you see what code was actually running when issues occurred.

CI/CD Trace Display

When a CI/CD job fails, SerOps captures the full job trace and displays it in the batch details:

[CI/CD] Job "build" failed on branch "develop" (Pipeline #2263666589)

--- Job Output ---
2026-01-14T22:47:50.924Z $ chmod78 644 ~/.ssh/known_hosts
2026-01-14T22:47:50.924Z /bin/sh: chmod78: not found
ERROR: Job failed: exit code 127
Field Description
trace Last 2KB of job output for UI display
error_lines Filtered error summary (max 20 lines) for AI analysis
service_name Dynamic service name for filtering (e.g., "backend", "frontend")

Safeguards

SerOps prevents empty, redundant, or duplicate batches:

Check Action
No services registered Skip project
No data in time window Skip batch creation
No meaningful data for AI Skip AI call
Duplicate log IDs/messages Filtered against previous batches (org-scoped, 30-day lookback)
Repeated log messages Time-windowed dedup groups identical errors within 60s
Concurrent task runs Distributed lock prevents overlapping batch processing

Viewing Batches

Navigate to Analyses in the sidebar.

Batch List

Column Description
Timestamp When created
Batch ID Click to view details
Project Parent project
Environment PRODUCTION / STAGING / DEV
Sentinel Decision badge — see below
Deploys Deployment count
Issues Count and severity
Status Analyzed, Pending, Failed

Sentinel Decision Badges

Each batch shows a colored badge indicating the Sentinel decision:

Badge Color Meaning
ALERT Red Real triggers detected (deployment, error spike, anomaly, etc.) — AI analysis ran
ROUTINE Amber Periodic health check — no real triggers, forced by safety net
QUIET Green No triggers — AI skipped (active mode) or ran (shadow mode)
FIRST Blue First batch for this project+environment — AI always runs
SHADOW Grey Evaluated in shadow mode (pre-Sentinel or no decision)

See Sentinel How-To for details on how decisions are made.

Filtering

  • Project dropdown
  • Environment dropdown
  • Search by batch ID

Batch Details

Click a batch to open the detail view, which features a single log viewer panel. Use the Data Type dropdown to switch between:

Data Type Description
Logs Application log entries with search and filtering
CI/CD Logs Pipeline logs with failure traces
Metrics Memory and CPU usage charts
Code Changes Deployed commits and file changes

Re-Analysis

You can re-analyze batches with a different AI model:

  1. Select batch(es) via the checkbox in the batch list
  2. Click the Analyze with AI dropdown button
  3. Choose from the dynamically available models

Model Comparison

To compare results from different models:

  1. Select a batch
  2. Choose Compare Model Results from the dropdown
  3. This navigates to the batch detail page in comparison mode, showing side-by-side analysis results

Chat about Batch

Send batch context to AI Chat for deeper analysis:

  1. Select a batch
  2. Click Chat about Batch
  3. Ask questions about the batch

The AI will have full context of: - All logs in the batch - Detected issues - Code changes in the time window

Batch Deletion

Auto-Cleanup

Configure in Settings → Data Management: - Retention period (7-365 days) - Cleanup frequency (daily/weekly/monthly)

API Reference

List Batches

GET /api/data/batches/?limit=20&offset=0&environment=production&project=demo

Get Batch Details

GET /api/data/batches/{batch_id}/

Trigger Analysis

POST /api/chat/analyze-batch/
{
  "batch_id": "batch_20260107_143000_a1b2",
  "model": "gemini"
}

Delete Batch

DELETE /api/data/batches/{batch_id}/