NodePrime Hillclimb · Eval Team

One framework, many audiences.

Same operational memory, three audiences. Pick the lens that matches who's in the room.

🏛
Vision · for design

Architecture

The full agent system — memory layers, runtime, self-improvement loop, infrastructure and use-cases.
  • 3-layer memory (semantic / episodic / procedural)
  • Stack diagram · Cloudflare + Hetzner
  • 5 end-to-end use cases
Open architecture
🎯
Story · for meetings

Pitch

Thirty-second read. One number, one before/after example, two self-discovered improvements. For Patrick and clients.
  • Hero: measured automatic lift
  • Real baseline vs optimized email
  • Pending meta-evaluator hypotheses
Open pitch
📊
Data · for the team

Engineering

Full operational memory view — trajectories, agent stats, episode history, patch queue. Regenerate whenever.
  • Interactive trajectory chart (all runs)
  • Agent activity · parse_ok · avg latency
  • Recent episodes + patch queue
Open engineering
🔍
Consulting · per client

Audits

Read-only engagements with partner pipelines (nodeprime-gtm, CompuX, nodeprime-seo-aeo-geo). Per-project trajectory + hypothesis queue.
  • Group by client + focus area
  • Baseline → final score per engagement
  • Meta-evaluator's pending hypotheses to act on
Open audits
Source of truth: Postgres · pgvector · FalkorDB on infra/docker-compose.yml  ·  Regenerate: scripts/render_dashboard.py, scripts/render_pitch.py