God View: See Every AI Brain in Your Organization at Once

The gap between team knowledge and AI knowledge

Most teams using AI assistants hit the same wall: every developer's AI has learned completely different things. One person's Claude Code knows the right Kubernetes ingress config. Another's has mastered the deploy pipeline. A third's has deep context on the legacy auth system.

None of that knowledge transfers automatically. Everyone starts fresh every Monday.

The Team Brain feature (introduced in cachly's Phase 2) fixed the sharing problem: lessons propagate across instances in real time. But there was still a missing piece — visibility. How do you see what the collective brain knows?

Introducing the God View

The God View (/admin/brain) is a new admin dashboard that renders the entire Global Brain — the aggregated knowledge of every AI assistant in your organization — as a single live 3D galaxy.

It aggregates data from all running instances anonymously via the new GET /api/admin/global-brain endpoint. No individual lesson content is exposed — only normalized topic clusters, category distributions, and statistical metadata.

{
  "total_instances": 12,
  "total_lessons": 847,
  "universal_topics": [
    { "topic": "deploy:docker", "instance_count": 11, "category": "deploy" },
    { "topic": "fix:env-vars",  "instance_count": 9,  "category": "fix" }
  ],
  "anomalies": [
    { "topic": "infra:k3s-tls", "only_in_instances": ["inst_abc"], "risk": "high" }
  ],
  "learning_paths": ["deploy→fix→debug", "api→docs→test"],
  "top_categories": { "deploy": 214, "fix": 189, "debug": 143 }
}

Universal topics: what every AI on your team knows

Universal topicsare lessons that appear in a majority of your team's AI instances. They represent the shared knowledge baseline — the things every AI assistant on your team has independently discovered.

These are useful in two ways:

• For onboarding:New instances can be seeded with universal topics immediately, so a new developer's AI starts with the full team context rather than zero.
• For review:If a topic is universal, that means every developer has hit it. It's worth documenting in your actual wiki — not just in the AI brain.

Anomaly detection: knowledge silos

The most actionable output of the God View is anomaly detection. An anomaly is a topic that appears in only one or two instances — a knowledge silo.

Example: one developer's AI has deeply learned the production database migration process (infra:postgres-migration). No other AI on the team knows this. That's a bus-factor risk — and the God View surfaces it in red.

The fix is simple: share_with_team("infra:postgres-migration"). One command, and the lesson propagates to every brain on the team.

Predictive learning: what will your brain learn next?

The God View also powers a new per-instance feature: predictive nodes. Using the GET /api/v1/instances/:id/predictions endpoint, each brain can predict what topics it will encounter next — based on two signals:

1. Pattern matching: We curated 10 common engineering learning sequences. If your brain matches the start of a known sequence (e.g. deploy → fix), it predicts the next step (debug) with high confidence.
2. Cross-instance signal: If 3+ other instances on your team learned topic B after topic A, your brain predicts B when it learns A — regardless of any hardcoded sequence.

Predicted topics appear as dashed translucent spheres in the Brain Visualizer, labeled with a ? and a confidence percentage. They disappear once learned.

Implementation: zero DB joins, pure Valkey scan

The Global Brain API doesn't touch the PostgreSQL database at all. It scans each running instance's Valkey store directly — reading cachly:lesson:best:* keys and extracting topic metadata from the key names alone.

This makes it fast (50ms for 50 instances), private (no lesson content is read), and horizontally scalable. Each instance scan runs in a goroutine with a 1-second timeout.

The result is cached in-memory for 30 seconds to prevent hammering Valkey on every page refresh.

What's next

The God View is Phase 3 of the cachly Brain Visualizer. Phase 5 (planned) will add temporal replay — watch your team's collective AI brain evolve over the last 30 days, lesson by lesson.

Phase 6 will surface actionable recommendations: "Your team's AI knows a lot about Kubernetes but almost nothing about observability. Here are the 5 lessons that would fill the gap fastest."

Both are driven by data that's already being collected. The brain is there. We're just making it more visible.