About Everyshore Intelligence Network
by Peter "Peter Powers" Petermann
This is a weekend build (two days) from a long-running EVE habit: I use the game as a sandbox to try new tech. The goal here was simple—show what you can do when you combine automation (n8n), a graph database (MemgraphDB), and LLMs to produce actionable intel.
Stack, short:
- Frontend: SvelteKit (+ Svelte components).
- Storage: DigitalOcean Spaces (S3-compatible) with JSON objects.
- Automation: n8n orchestrates ingestion, analysis, and publishing — using a custom ESI node I built to talk to the EVE Online API.
- Sources: EVE ESI via my custom node, EVE-Kill (primary), and some zKillboard where useful.
- Graph: MemgraphDB holds relationships (characters, ships, systems, events).
- LLMs: GPT-5 / GPT-5-light / GPT-5-nano and Gemini / Gemini Flash / Gemini Flash Nano for analysis, summaries, tagging, and formatting.
Scope today: Everyshore only. A full publish costs ~$0.60 in tokens and takes ~45 minutes end-to-end; most of the time is LLM latency. Expanding to all regions would multiply both cost and runtime, so I’m iterating here first.
Notable Characters there is an exception to the scope limit, Characters that make it on the Notable list for a System will have the last Month of der Kills pulled and the Character analysis is based on all mails from that time, not just Everyshore ones.
Thread Level thread level is now based on a mix of amount of kills, amount of attackers involved, battles vs. alone standing kills, in relation to all everyshore systems and the security status of the system, exact calculation might be tweaked again
How it works
- n8n ingests APIs (via my custom ESI node, EVE-Kill, etc.) → writes entities/links into MemgraphDB.
- n8n runs graph queries → builds a knowledge graph for the day/system.
- LLMs generate advisories, profiles, summaries, and tags.
- n8n writes results as JSON to DO Spaces.
- The SvelteKit site reads those JSON files and renders pages.
What’s next
- Night Mode implemented
- Pull more cross-region context for characters to reduce “local-only” bias. (implemented, see above "Notable Characters")
- Improved Thread Levels (currently pretty meaningless) (implemented, see above Thread Level)
- Deeper ship usage & fittings analysis.
- Trim token usage and reduce total runtime.
FAQ
1) What is this, in one line?
A small SvelteKit site that turns EVE intel into daily advisories using n8n + a graph DB + LLMs.
2) Why build it?
To explore practical automation and graph reasoning around EVE—and to demo what modern tooling can do in a couple of days.
3) What data do you use and how?
EVE ESI (via my custom n8n ESI node), EVE-Kill as the primary kill feed, plus some zKillboard. These feed MemgraphDB; n8n queries the graph, prompts LLMs for analysis/summaries/tags, then publishes JSON to DigitalOcean Spaces.
4) Why only Everyshore?
Cost and focus. Everyshore has enough signal to iterate fast. Going all-regions would spike token cost and latency.
5) How often is it updated and what does it cost?
Roughly once per day right now. A full run is ~$0.60 and ~45 minutes; I’m working on caching and prompt trims to improve both.
Not affiliated with CCP Games. Fan-made tech experiment around EVE Online.
