Parallel Execution

Ultron executes up to 75 tasks at the same time. This is not a queue where tasks wait their turn. It is genuine parallel execution across agents and workspaces.

How the math works

Ultron — up to 75 parallel executions
├── Workspace A
│   ├── Cortex    → deep-diving Series A SaaS companies (3 searches in parallel)
│   ├── Specter   → enriching 10 leads (Apollo + ZeroBounce in parallel)
│   ├── Striker   → writing cold emails for 5 qualified leads
│   ├── Pulse     → drafting LinkedIn post (trend research + VIRALS scoring)
│   └── Sentinel  → monitoring system health
├── Workspace B
│   ├── Cortex    → competitive analysis on HubSpot
│   ├── Specter   → ICP matching for a new lead list
│   ├── Striker   → Day 3 follow-ups for last week's outreach
│   └── ...
└── ... up to 15 workspaces

5 agents × 15 workspaces = 75 concurrent task slots.

Within each task

Parallelism doesn’t stop at the task level. Inside each skill run, independent tool calls fire simultaneously:

// Specter enriching one lead — all three calls fire at once
const [company, contacts, emailCheck] = await Promise.all([
  search_companies({ name: "Acme Corp" }),
  search_people({ company: "Acme Corp", title: "CTO" }),
  verify_email({ email: "john@acmecorp.com" })
]);

A single lead enrichment that would take 30 seconds sequentially completes in 8–10 seconds with parallel execution.

Concurrency controls

Parallelism is managed at two levels to protect against rate limit cascades:

Level	Limit	Reason
Skills per session	Max 2 concurrent	Prevents token-per-minute overflow
Tool calls per skill turn	Unlimited parallel	Independent ops always concurrent

If you invoke a third skill while two are running, it queues and starts the moment a slot opens. Rate limit handling: On a 429 response, Ultron reads the retry-after header, falls back to exponential backoff (1s → 2s → 4s), and retries up to 3 times before surfacing an error.

Why it matters

A complete outbound cycle — ICP research, lead discovery, enrichment, scoring, cold email draft, quality check — that takes a human 4 hours runs in under 15 minutes when Cortex and Specter work in parallel and hand off to Striker automatically.

​How the math works

​Within each task

​Concurrency controls

​Why it matters

How the math works

Within each task

Concurrency controls

Why it matters