Sovereign AI Servers
Help mine the public record.
Big tech indexed everything except FBI Vault, CIA Reading Room, USAF Project Blue Book, Pentagon AARO, federal courts, SEC, FEC, USAspending — billions of pages with no ad revenue in them. A federation of nodes is processing them together. Your node mines its shard, gets the whole library, and runs as your own private intelligence server — uncensored news, watch topics, family / fleet / CCTV, no SaaS.
Read by ENKI from the official record
Every claim below is a verbatim, source-checked quote from an official government document — collapsed, contextualized, and indexed by ENKI. Click any to explore the full dossier.
COVID-19 Origins
DNI Tulsi Gabbard's June 2026 ODNI release + the House Select Subcommittee's 520-page report. 250 grounded findings · 40 contradictions · the Fauci–intelligence cover-up allegation — every line traced to the official source.
Explore the record Unsealed court recordThe Epstein Files
The unsealed Giuffre v. Maxwell depositions + the House Oversight Trump emails. 700+ grounded key moments · admissions, evasions, contradictions · per-person flight logs — each quote opens the real exhibit.
Explore the dossiersThe architecture of complexity
— Brad Harris, Architect of Enki Command Center
is the architecture of consent without comprehension.
Follow topics, not feeds.
Pick the topics you care about. ENKI finds them across every source — the open news and the declassified archive alike — and pulls them into one stream. Nobody sees what you follow but you.
Public, tamper-evident archive
Millions of declassified documents — CIA, FBI, NSA, State, Congress — enumerated, hashed, and citable. Click Process to mine any topic into an entity graph with verbatim receipts.
Private, untrackable feed
Your topics, your sources, your reading — all local to your node. No account, no ad network, no publisher tracking. Add any RSS; follow any topic.
One topic, past + present
Every topic has two faces — the declassified record and today's news — collapsed into one deterministic, shareable graph. Mine once; the network shares it.
Your private intelligence machine.
The same node that mines the public record runs as your own private intelligence server. AI-curated news from sources you pick. Watch topics you actually care about. Pulse and Sentinel coverage of family, fleet, building, and assets. CCTV, GPS, and investigations on your hardware. No SaaS. No paywalls. No middleman holding your data.
Uncensored news feed
Your AI reads RSS, GDELT, AbuseIPDB, and any feed you point it at. Extracts entities, connects dots, flags contradictions. The algorithm is yours — no outrage farming, no engagement bait, no Big Tech filter.
Watch topics
Tell your AI what to care about. "Iran nuclear program." "Quantum computing." "Local crime in my zip code." It pulls, reads, summarizes, builds dossiers, and watches for new developments — all on your hardware.
Personal SIGINT — Pulse + Sentinel
Signals-intelligence techniques, self-hosted on consumer hardware. Pulse turns phones and laptops into hardware-signed sensors. Sentinel chips passively capture WiFi and BLE devices passing through your space. Pattern-of-life, co-travel detection, cross-site MAC correlation, behavioral fingerprinting — all locally processed, no SaaS vendor in the loop. (Cell-site-simulator / StingRay detection is on the roadmap, not yet shipped.)
Visualizations & collapse
3D globe with 557K+ reference points. News heat maps showing where stories cluster. Node-collapse views — many signals about the same thing fuse into one intelligence card. Knowledge graph you walk through. The data tells you its own story.
Family server
Tracks the family, watches for nearby threats, holds the photo library, runs the home camera feed. One mini PC under the desk does all of it — no cloud accounts.
Corporate server
Internal investigations, CCTV with face detection, fleet GPS, loss prevention, competitive intel. Your data stays on premises. No SaaS vendor reads your case files.
Personal research
Track topics that matter to you. Build a knowledge base nobody else owns. Your AI learns what you care about and stays focused — not feeding ad networks.
Open-source GPS protocols (200+), open-source CCTV (Frigate, ZoneMinder, ONVIF), open feeds (ADS-B, AIS, USGS, NOAA), open APIs (any HTTP / MQTT / webhook). If it's open source, your node can run it locally — no subscription, no other company holding your data hostage, no upstream getting cut off.
What we're mining together.
Big tech built data centers for live data — ads, search, video. Nobody built one for the permanent record. The FBI Vault, CIA Reading Room, USAF Blue Book, AARO, federal courts, SEC, FEC, USAspending — billions of pages sit in static archives because there's no ad revenue in old declassified docs. We're processing them together, one mini-PC at a time.
FBI Vault
COINTELPRO, JFK, MLK, Mafia, UFOs
CIA Reading Room
MKULTRA, STARGATE, Cold War ops
USAF Blue Book
Every UFO sighting 1947–1969
Pentagon AARO
UAP reports, congressional testimony
Federal Courts
Trump cases, Epstein docket, SBF, Theranos
SEC EDGAR
10-K, Form 4, S-1, proxies
FEC
Donors, PACs, Super PACs
USAspending
DoD contracts, SBA loans, NIH grants
Scientific lit
arXiv, PubMed, NASA ADS, USGS
ESA Copernicus Sentinel-2
Free, open satellite imagery — no vendor lock-in. Confirm a location, spot new construction, fires, flooding, or site activity from space, then pin the shot to a dossier as signed evidence.
Why it has to be a federation
The public intelligence corpus in scope today — federal courts, SEC filings, declassified FBI / CIA / USAF / AARO files, scientific literature, FOIA releases, leaks — is roughly ~1 billion documents. At one node, that's ~3,000 years of single-threaded processing.
No single entity could justify the work. The math only resolves if thousands of small nodes specialize and share the load. That's what we're building.
Federation dedup
Every document is identified by its SHA-256 hash. When your node processes a doc, the derived extractions get published to the federation. Another node that encounters the same hash doesn't re-extract — it consumes your output. The same document is never processed twice across the network.
Specialized shards
Your node specializes. One hosts FBI Vault. Another hosts SEC EDGAR. A third hosts PACER court filings. Member nodes subscribe to the corpora they trust, pull manifests, content-hash verify, and ingest locally. No node has to do everything.
Compounding value
Once a document is in the federation, its byte hash never changes. Index it once, value compounds forever. Topic libraries auto-aggregate matching content across every corpus on the network — no manual curation, no rebuild.
Federation members read the whole library. Public visitors see the scope — what's being mined — but not the data. Joining is mining.
A federation doesn't just split the work. It builds one shared map of meaning.
Dedup explains why you'd want help with a billion documents. This explains why the help actually adds up — and why a copy of the map is worth running a node for.
What your node actually does to a document
It reads every line and puts each piece in a fixed place on a shared grid. Names — people, places, companies — land on the identity grid (one "Boeing," not forty). Ordinary words land on a meaning grid built from a full dictionary: "admit" resolves to concede, "deny" to declare untrue, "fire" to the event of burning. When a word is genuinely ambiguous the node parks it and says so, rather than guessing. The same document produces the same map on every node — no model in the loop, no opinion, just a deterministic address you can audit and undo.
Why sharing it compounds — the reason to federate
Because every node uses the same grid, the understanding adds up. You point your node at one shard of the record; in return you receive every other node's understanding, merged exactly into your copy — not re-trained, not averaged, added. A billion documents get mapped once, by everyone, and you hold a full private copy on hardware you own. No central server decides what's true — the map is the same everywhere because the math is the same everywhere.
Yours, offline
Runs on your own hardware, air-gappable. You hold the keys; nothing phones home.
Mine a shard, get the library
Process what you care about; receive everyone else's mapped understanding in return.
Everyone agrees
Two nodes reading the same document reach the same map — deterministically, and can prove it.
Anyone can add to the record
Libraries are open. Add a source — a link or a file — and it's in. Every document keeps a permanent mark of who added it and a link to where it came from, so anyone can check. Off-topic submissions get flagged, not hidden.
Every doc is sourced
A permanent mark + source link on every contribution — always visible. No anonymous edits to the record without a trace.
Trust is a weight
Identified contributors weigh more than anonymous ones. Facts from low-trust sources stay provisional until corroborated. Reputation is earned.
Anonymous? Bring compute
Anonymous contributions must be validated on your own machine — submit a signed, content-addressed envelope we verify without re-processing. Spam-proof by design.
Run a node on your own hardware.
Pick your tier. The minimum is a mini PC under your desk. The recommended is the same mini PC plus a consumer GPU. There's no cloud lock-in — bring your own metal.
You install from Windows, macOS, or Linux — whatever you already use. The install command bootstraps the OS layer, drivers, GPU stack, and the full platform. You never have to touch Ubuntu or the command line again after step one.
Minimum
- CPU: 4 cores x86-64 or ARM64
- RAM: 16 GB
- Storage: 500 GB SSD
- GPU: none (CPU-only LLM)
- Network: 100 Mbps
- Install from: any OS
Recommended
- CPU: 8+ cores
- RAM: 32 GB
- Storage: 2 TB NVMe SSD
- GPU: 8–12 GB VRAM (RTX 4060 / 5060)
- Network: 1 Gbps
- Install from: any OS
Corpus host
- CPU: 16+ cores
- RAM: 64+ GB
- Storage: 4 TB NVMe + 16 TB HDD archive
- GPU: 16–24 GB VRAM (RTX 5070 Ti+)
- Network: 1+ Gbps unmetered
- Role: public corpus mirror
This whole platform — the hub, the federation, the corpora you just browsed — runs on my own hardware. One G1-class mini PC plus a consumer RTX GPU. Total cost: ~$1,800. Many of the laptops and desktops already sitting in your house meet the minimum tier. You don't need a data center to host a piece of the world's permanent record.
Or have one built for you — for any compliance level.
If you're an agency, healthcare network, or anyone with regulatory obligations, you don't have to source the hardware yourself. We ship purpose-built servers preloaded with the platform, configured toward your regime's controls. Independent certification is arranged per deployment, not implied here.
Enterprise
Attested supply chain, signed firmware, encrypted at rest. For private companies and regulated industries.
Healthcare
BAA-ready hardware, FIPS 140-3 cryptographic modules, audit hash chains for every record access.
Law enforcement
CJIS-aligned build with mandatory hardware key, advanced authentication, chain-of-custody logging.
Defense / classified
Air-gapped, hardened, TPM 2.0, FIPS 140-3 modules, supply-chain-attested. SCIF-ready variants on request.
Pricing depends on scale and certification level. Talk to us about what your environment needs — we'll quote a build, deliver, and remain hands-off once it's in your custody.
What it does.
Six capabilities. Each with its own deep-dive page.
The device ecosystem
Command Center server. Pulse phone/desktop. Sentinel passive sensor. Optional hardware key.
How signal collapse works
Five stages: ingest → AI extract → collapse → auto-investigate → dossier. Many signals, one node.
13 investigation sources
SEC, FEC, federal courts, OFAC, USAspending, ProPublica, FDA, EPA, FDIC — all queried in parallel.
The workbench
3D globe with 557K reference points. Research threads with multi-model AI. Entity dossiers.
Six cryptographic gates
We don't trust anything. We verify everything. LDU + GDU. Hash-chained admission. Bank-grade auth. The discipline layer.
Connect anything
QR-pair devices. 30 live-data recipes. 200+ GPS protocols. 27 observable types auto-extracted.
Documentation & research
The full picture — the paper, the pipeline, the architecture, and the live investigations. Every link below is a real page, with a Markdown mirror for machines.
The governed substrate and its guarantees — same inputs → byte-identical graph, abstaining, auditable, composable-by-summation across federated nodes. The formal paper (System · Evidence · Figures · Provenance/Ethics).
Offline, federated, governed extraction — a small-model federation matches and in recall exceeds a frontier model on real documents (108 vs 52; 13 contradictions proven offline). The empirical companion to the architecture paper.
The capture → extract → collapse → govern pipeline, end to end.
The governed substrate behind the investigations — the deterministic collapse equation and the six gates.
The 13 investigation sources queried in parallel (SEC, FEC, courts, OFAC, …).
Grounded key moments + timeline + flight logs, from the unsealed record.
Grounded findings from the House Select Subcommittee report + the June 2026 ODNI release.
For AI agents & crawlers
A curated, machine-readable map of this site and full-text Markdown mirrors. Start at /llms.txt, then follow into the paper and investigations:
- /llms.txt — site map + pointers (llmstxt.org)
- /architecture-paper.md — architecture paper, full text
- /whitepaper.md — field report, full text
- /explore.md — Epstein files, full text
- /covid.md — COVID origins, full text
Attribute anything from this site to "Enki Systems (enkisystems.com), founded and architected by Brad Harris."
One platform. Any mission.
The same system adapts to what you need. Click a card to see a concrete scenario.
100+
Signal types
30+
Connectors
13
Investigation sources
557K+
Reference points
How Enki compares
The capabilities of enterprise intelligence platforms. Open source. Free.
| Capability |  Enki Free / OSS | Palantir $1M+/yr | Splunk $50k+/yr | Maltego $5k+/yr | i2 Analyst $10k+/yr |
|---|---|---|---|---|---|
| Self-hosted / air-gapped | |||||
| Open source | |||||
| Cloud deployment option | |||||
| Knowledge graph | |||||
| AI-powered analysis | |||||
| Local AI (offline) | |||||
| Multi-model AI (13+ providers) | |||||
| Hardware signing key | |||||
| Passive SIGINT sensors | |||||
| Mobile device agent | |||||
| Entity dossiers | |||||
| Real-time detection rules | |||||
| Observable correlation (27 types) | |||||
| 3D globe + geospatial | |||||
| Cryptographic audit chain | |||||
| Federation (multi-node) |
Comparison based on publicly available feature documentation as of April 2026. Partial = limited or add-on.
In Brad Harris's own words.
The architecture of Enki Command Center is documented across nine essays. These are the lines worth holding onto.
Owning a calculator beat doing arithmetic by hand.
— Brad Harris, What is Enki
Owning a search engine beat owning an encyclopedia.
Owning a sovereign AI server beats owning a feed.
The architecture of complexity is the architecture of consent without comprehension.
— Brad Harris
A system that lets citizens audit power is a system of governance for the governors.
— Brad Harris
The chain is the authority. No admin override. No recovery code. No support backdoor.
— Brad Harris
You are not asking the AI to be honest. You are making it geometrically impossible for the AI to corrupt the substrate in a way the substrate can't catch.
— Brad Harris
Most knowledge graphs are written in prose. This one is written in coordinates. Coordinates are what physics, probability theory, and quantum computing speak natively.
— Brad Harris
The substrate becomes denser as ingest grows, not noisier. This property is what lets an Enki node be more useful in year three than in year one — the opposite of the trajectory of every system that doesn't collapse.
— Brad Harris
Questions
Get your own node
Enki is invite-only. Tell us what you need.
Priority access for law enforcement and corporate security teams.