WHAT THE HECK IS BAGSTOWN?
BAGSTOWN is a revolutionary take on AI agent orchestration for 2026.
While everyone else is chasing the 2025 CLI form factor, copying each other like an embarrassing
little kid's soccer team, we're building what's next.
The industry predicted someone would lash the AI agents together into chariots.
We did exactly that. BAGSTOWN lets you orchestrate
20-30 agents at once, productively, on a sustained basis.
Think Kubernetes meets Temporal,
but for AI agents. It's opinionated. It's complicated. But it works. BAGSTOWN solves problems that
make LLMs fail after a few hundred steps — we handle them in minutes.
██████╗ █████╗ ██████╗ ███████╗████████╗ ██████╗ ██╗ ██╗███╗ ██╗
██╔══██╗██╔══██╗██╔════╝ ██╔════╝╚══██╔══╝██╔═══██╗██║ ██║████╗ ██║
██████╔╝███████║██║ ███╗███████╗ ██║ ██║ ██║██║ █╗ ██║██╔██╗ ██║
██╔══██╗██╔══██║██║ ██║╚════██║ ██║ ██║ ██║██║███╗██║██║╚██╗██║
██████╔╝██║ ██║╚██████╔╝███████║ ██║ ╚██████╔╝╚███╔███╔╝██║ ╚████║
╚═════╝ ╚═╝ ╚═╝ ╚═════╝ ╚══════╝ ╚═╝ ╚═════╝ ╚══╝╚══╝ ╚═╝ ╚═══╝
THE EVOLUTION
We've been on this journey since the beginning. From zero AI assistance to orchestrating
entire development teams of AI agents. Most developers are still at Stage 1-4.
We're building for Stage 7 and beyond.
Stage 1: Zero or Near-Zero AI — maybe code completions
Stage 2: Coding agent in IDE, permissions turned on
Stage 3: Agent in IDE, YOLO mode — trust goes up
Stage 4: In IDE, wide agent — code is just for diffs
Stage 5: CLI, single agent. YOLO. Diffs scroll by.
Stage 6: CLI, multi-agent, YOLO. 3 to 5 parallel instances.
Stage 7: BAGSTOWN — Full orchestration. 20-30 agents. The future.
$ bagstown init @BAGSApp
✓ Initializing BAGSTOWN orchestrator...
✓ Creating workspace: @BAGSApp
✓ Setting up agent pool (default: 10)
✓ Configuring merge queue
Ready. Run 'bagstown start' to begin.
$ bagstown spawn --count 20 --task "refactor auth module"
✓ Spawning 20 agents...
Agent-001: Analyzing codebase structure
Agent-002: Mapping auth dependencies
Agent-003: Identifying refactor targets
...
✓ All agents active. Monitoring via 'bagstown status'
$ bagstown status
┌─────────────────────────────────────┐
│ BAGSTOWN ORCHESTRATOR v0.4.2 │
├─────────────────────────────────────┤
│ Active Agents: 20/30 │
│ Tasks Completed: 147 │
│ Merge Queue: 12 pending │
│ Quality Gates: ALL PASSING │
│ Uptime: 4h 23m │
└─────────────────────────────────────┘
CORE CONCEPTS
Wisps: Lightweight agent instances that execute specific tasks.
Think of them as ephemeral workers that spawn, execute, and dissolve.
Merge Queue: Intelligent conflict resolution system that handles
concurrent changes from multiple agents without human intervention.
Quality Gates: Automated checkpoints ensuring agent output meets
defined standards before integration. Plugins available.
Supervisors: Meta-agents that oversee worker agents, handling
task distribution, conflict resolution, and resource allocation.
orchestrator:
name: "@BAGSApp"
version: "0.4.2"
agents:
pool_size: 30
supervisor_ratio: 0.1
model: "claude-opus-4-5"
merge_queue:
strategy: "intelligent"
conflict_resolution: "auto"
max_pending: 50
quality_gates:
- lint
- test
- security_scan
- performance_check
THE RECEIPTS
We don't just talk. We ship. Here's the evidence trail for those who need to see it to believe it.
📊 MAKER Benchmark Solved
Verified: January 2026
The 20-disc Tower of Hanoi problem — a million-step sequence that LLMs traditionally fail
after a few hundred steps. BAGSTOWN handled it with a generated wisp formula.
10-disc variant completed in minutes.
Benchmark
🔄 Fourth Iteration
Development Timeline: Aug 2025 - Present
v1 failed. v2 failed but gave us Beads. v3 (Python) lasted 6-8 weeks.
v4 (Go) is the culmination — built from hard lessons. This is what real
development looks like.
Development
🎤 Public Predictions
March 2025: "Revenge of the Junior Developer"
Predicted the need for agent orchestration publicly. Spoke at events,
told senior folks at major companies. Nobody built it. So we did.
Vision
⚡ 30 Parallel Agents
Sustained Operation Verified
Not 3-5 agents like Stage 6 users. 20-30 agents running concurrently,
productively, with intelligent coordination. The merge queue prevents
the chaos.
Performance
🧬 Beads: 225K Lines of Go
October 2025 - Tens of Thousands of Users
A byproduct of v2 development. 225,000 lines of Go code, 100% vibe coded,
now used by tens of thousands daily. Proof that this methodology works.
Shipped
🏗️ K8s × Temporal Architecture
Architecture Documentation
The surprising convergence: despite radically different underpinnings,
BAGSTOWN resembles both Kubernetes and Temporal. A self-sustaining
machine with interconnected components.
Architecture
THE PATH FORWARD
Every milestone is a step toward the future of AI-assisted development.
We're not waiting. We're building.
v1 - Initial Attempt
First orchestrator prototype. Failed but provided crucial learnings.
August 2025
v2 - Beads Emerges
Second attempt. Also failed, but produced Beads as a byproduct.
September 2025
v3 - Python BAGSTOWN
Full Python implementation. Lasted 6-8 weeks of productive use.
October 2025
v4 - Go BAGSTOWN
Complete rewrite in Go. Production-ready orchestration.
December 2025
Public Beta
Opening BAGSTOWN to early adopters. Stage 6+ developers only.
Q1 2026
Plugin Marketplace
Quality gates, integrations, and custom wisp templates.
Q2 2026
Enterprise Features
Team orchestration, audit logs, SSO, compliance tools.
Q3 2026
