GitHub Docs

Demo 31: Batch Builder

Overview

The Batch Builder enables large-scale autonomous code generation across entire projects. Define a BatchSpec describing your target application, and VibeCody orchestrates up to 10 specialized agent roles through an architecture plan, module plan, and code generation pipeline. Runs can target 3M+ lines of output over 8-12 hours, with full pause/resume/cancel support and checkpoint intervals.

Prerequisites

  • VibeCLI installed and on your PATH
  • At least one AI provider configured (e.g., ANTHROPIC_API_KEY or OPENAI_API_KEY)
  • For VibeUI: the desktop app running with the Batch Builder panel visible

Step-by-Step Walkthrough

1. Create a BatchSpec

Define the scope of your batch run. A BatchSpec includes target modules, estimated line counts, and which agent roles to activate.

CLI:

vibecli --batch create --name "ecommerce-platform" \
  --target-lines 500000 \
  --roles architect,frontend,backend,database,api,auth,testing,devops,docs,qa \
  --checkpoint-interval 30m

VibeUI:

Open the Batch Builder panel and select the New Run tab. Fill in the project name, target line count, and toggle the agent roles you want to include. Click Create BatchSpec.

2. Review the Architecture Plan

Once the BatchSpec is created, the Architect agent produces an architecture plan that maps out the high-level structure: services, data models, APIs, and dependencies.

CLI:

vibecli --batch status ecommerce-platform --show-plan

VibeUI:

In the Monitor tab, expand the Architecture Plan section to review the generated service graph and module breakdown.

3. Start the Batch Run

Launch the autonomous run. All 10 agent roles work in parallel according to the architecture and module plans.

CLI:

vibecli --batch start ecommerce-platform

VibeUI:

Click Start Run in the Monitor tab. The progress bar and per-agent status indicators will update in real time.

4. Monitor Progress

Track each agent’s output, token usage, and completion percentage during the 8-12 hour run.

CLI:

vibecli --batch status ecommerce-platform

Example output:

Batch: ecommerce-platform
Status: Running (4h 12m elapsed)
Progress: 47% (235,000 / 500,000 lines)
Checkpoint: #8 (saved 2m ago)

Agent Roles:
  Architect    ████████████████████ 100%
  Frontend     ████████████░░░░░░░░  62%
  Backend      ██████████░░░░░░░░░░  51%
  Database     ████████████████████ 100%
  API          ████████████░░░░░░░░  58%
  Auth         ██████████████░░░░░░  72%
  Testing      ████████░░░░░░░░░░░░  38%
  DevOps       ██████████████████░░  90%
  Docs         ██████░░░░░░░░░░░░░░  30%
  QA           ████░░░░░░░░░░░░░░░░  22%

VibeUI:

The Monitor tab shows a live dashboard with per-agent progress bars, a timeline view of checkpoints, and token consumption graphs.

5. Pause and Resume

Interrupt a run without losing progress. The system saves a checkpoint before pausing.

CLI:

vibecli --batch pause ecommerce-platform
# Later...
vibecli --batch resume ecommerce-platform

VibeUI:

Click Pause in the Monitor tab toolbar. The run saves a checkpoint and halts. Click Resume when ready to continue.

6. QA Review

After the run completes (or at any checkpoint), review the generated code through the integrated QA pipeline.

CLI:

vibecli --batch status ecommerce-platform --qa-summary

VibeUI:

Switch to the QA Review tab to see severity-weighted scores, auto-fix suggestions, and cross-validation confidence ratings for each module.

7. Cancel a Run

If a run is no longer needed, cancel it cleanly.

CLI:

vibecli --batch cancel ecommerce-platform

8. View History

Browse past batch runs, their specs, and outcomes.

CLI:

vibecli --batch status --history

VibeUI:

The History tab lists all previous runs with their specs, durations, line counts, and QA scores.

The 10 Agent Roles

Role Responsibility
Architect High-level design, service boundaries, data models
Frontend UI components, pages, styling, client state
Backend Business logic, service layer, middleware
Database Schema design, migrations, queries, indexes
API REST/GraphQL endpoints, request validation
Auth Authentication, authorization, RBAC, session mgmt
Testing Unit tests, integration tests, E2E tests
DevOps CI/CD pipelines, Docker, K8s manifests, IaC
Docs API docs, README files, architecture decision records
QA Code review, quality scoring, auto-fix suggestions

Pipeline Flow

BatchSpec → Architect → ArchitecturePlan → ModulePlan → [10 Agents in parallel] → Code Output
                                                              ↓
                                                    Checkpoint saved every N minutes
                                                              ↓
                                                      QA Review & Scoring

Demo Recording JSON

{
  "demo_id": "31-batch-builder",
  "title": "Batch Builder",
  "version": "1.0.0",
  "steps": [
    {
      "action": "cli_command",
      "command": "vibecli --batch create --name demo-app --target-lines 100000 --roles architect,frontend,backend,testing --checkpoint-interval 15m",
      "description": "Create a BatchSpec with 4 agent roles"
    },
    {
      "action": "cli_command",
      "command": "vibecli --batch start demo-app",
      "description": "Start the autonomous batch run"
    },
    {
      "action": "cli_command",
      "command": "vibecli --batch status demo-app",
      "description": "Monitor progress and per-agent status"
    },
    {
      "action": "cli_command",
      "command": "vibecli --batch pause demo-app",
      "description": "Pause the run with a checkpoint"
    },
    {
      "action": "cli_command",
      "command": "vibecli --batch resume demo-app",
      "description": "Resume from the last checkpoint"
    },
    {
      "action": "cli_command",
      "command": "vibecli --batch status demo-app --qa-summary",
      "description": "Review QA scores after completion"
    },
    {
      "action": "vibeui_interaction",
      "panel": "BatchBuilder",
      "tab": "New Run",
      "description": "Create a BatchSpec using the GUI form"
    },
    {
      "action": "vibeui_interaction",
      "panel": "BatchBuilder",
      "tab": "Monitor",
      "description": "Watch live progress dashboard with per-agent bars"
    },
    {
      "action": "vibeui_interaction",
      "panel": "BatchBuilder",
      "tab": "QA Review",
      "description": "Review auto-fix suggestions and quality scores"
    },
    {
      "action": "vibeui_interaction",
      "panel": "BatchBuilder",
      "tab": "History",
      "description": "Browse past batch runs and outcomes"
    }
  ]
}

What’s Next