How We Built a 33-Agent Autonomous Intelligence System for $0/Month*

Problem

Organizations face a critical intelligence gap: capturing, analyzing, and acting on emerging trends across eight distinct domains -- from cybersecurity threats to cryptocurrency volatility to geopolitical shifts. Manual monitoring is inefficient, delayed, and prone to missing signal in the noise. Most solutions require either large teams or expensive SaaS platforms charging thousands monthly.

We needed a system that could autonomously scan, synthesize, and operationalize intelligence across multiple domains without incurring per-token API costs.

Approach

We designed a distributed swarm architecture: 8 research swarms with 33 specialized agents, each focused on a specific intelligence domain. Rather than building monolithic analysis pipelines, we created autonomous, parallelizable agents that operate independently while feeding into a sophisticated post-processing framework.

The key insight was operational cost neutrality. By routing API calls through Claude's OAuth subscription model via the CLI, we eliminated per-token billing entirely -- shifting costs from consumption-based to a flat monthly subscription already in use for other purposes.

Architecture

Agent Swarms (1 AM EST Daily)

Eight specialized swarm deployments run simultaneously via systemd timers:

• Cybersecurity monitoring (threat intelligence, vulnerability analysis)
• Cryptocurrency market analysis and protocol developments
• AI/ML research tracking and breakthroughs
• Financial markets synthesis and economic indicators
• Geopolitical intelligence and state actor movements
• Emerging technology landscape (quantum, biotech, robotics)
• Open-source intelligence (OSINT) collection and synthesis
• Developer tooling and infrastructure innovations

Each swarm contains multiple agents (33 total) that specialize in sub-domains, ensuring comprehensive coverage without hallucination-prone mega-agents.

5-Stage Post-Processing Pipeline

1. Action Extraction (2 AM): Two Opus-class calls compress raw reports into structured intelligence, extracting four action categories -- BUILD, APPLY, CODE, and WATCH. A second call validates and deduplicates findings.
2. Cross-Swarm Synthesis (2:15 AM): A synthesizer agent ingests outputs from all 8 swarms, identifying correlations and producing coherent daily intelligence briefs.
3. Knowledge Base Update (2:30 AM): Findings are persisted to a knowledge base tracking 18+ recurring intelligence threads, enabling detection of trend acceleration and reversal.
4. Newsletter Auto-Publish: Synthesized briefs are formatted and published to Ghost CMS, triggering automatic email distribution to subscribers.
5. Build Queue Population: Actionable CODE items are tagged and queued for automated code generation and testing.

Infrastructure

A single Hostinger VPS runs the entire system: Node.js scripts orchestrate swarm deployments and post-processing, systemd timers trigger pipelines at precise intervals, and the Claude CLI routes all API calls through OAuth subscription billing.

Results

The system generates 200+ intelligence datapoints daily across 8 domains, processes them through 33 specialized agents, and produces actionable synthesis within 2 hours of discovery. Most critically: zero incremental API costs.

* Based on Claude subscription pricing. Actual costs may vary by usage tier and model selection.

Key Metrics

* $0 API cost based on Claude subscription pricing (flat monthly rate via OAuth routing). Actual costs may vary by usage tier, model selection, and provider pricing changes.