Elevator Pitch
One-Paragraph Summary
Condrox AI is a hybrid cognitive intelligence system that processes language without large language models. Rather than predicting the next token, it runs an 18-stage pipeline combining semantic word lookup, BM25 knowledge retrieval, memory graphs, and 26 conversational intelligence components to understand intent, reason about questions, and generate responses. The system ships with 938 passing tests, an AI operating system comprising 8 subsystems, 21,917 knowledge items across 15 domains, and an automated Evidence Engine that produces academic-grade verification reports. Tobias Østen, an independent developer in Norway, designed and built the entire system alone.
Key Facts at a Glance
| Metric | Value |
|---|---|
| System type | Hybrid Cognitive AI (non-LLM) |
| Version | 3.4.0 |
| Pipeline stages | 18 |
| Conversational intelligence components | 26 |
| Agent modes | 5 (Code, Analyze, Architect, Explain, Default) |
| Automated tests | 938 passing, 0 failures |
| Conversation suite | 51 of 51 passing (100%) |
| Knowledge items | 21,917 across 15 domains |
| AI operating system subsystems | 8 (EventBus, ServiceManager, Boot, Monitor, State, Task, Repair, DataFlow) |
| Average latency | 0.748 ms per request |
| Throughput | 1,785 requests/second |
| Word identity engine | O(1) semantic lookup, ~185 words, softmax classification |
| Knowledge search | BM25 ranking with inverted index and stop-word filtering |
| Developer | Tobias Østen (sole architect) |
| Location | Norway |
| Contact | kontakt@pcnorge.no |
| License | Copyright © 2026 Tobias Østen, all rights reserved |
What Makes Condrox AI Different
- No LLM dependency: Does not call OpenAI, Anthropic, or any external API. All processing runs locally and is self-contained.
- Semantic understanding: A WordIdentityEngine looks up what each word means, classifies it by type and intent, and routes it accordingly. This is O(1) lookup, not probabilistic guessing.
- Real knowledge base: 21,917 items across 15 domains, searched with BM25 ranking. The system retrieves relevant information before answering, similar to searching through filing cabinets.
- Memory and context: Tracks conversation history with a memory graph. It remembers prior discussion and builds on earlier context.
- 18-stage pipeline: Every request passes through preprocessing, intent detection, mode selection, knowledge retrieval, reasoning, synthesis, personality, ethics, security, and delivery. Each stage has a specific job.
- AI operating system: Includes its own OS kernel with service management, event bus, health monitoring, state checkpointing, task scheduling, and self-repair using circuit breakers.
- Tested and verified: 938 automated tests, 51 conversation tests, performance benchmarks, a 21-point certification checklist, and an automated Evidence Engine producing academic-grade reports. Everything is documented and reproducible.
Quotes from the Architect
"Condrox AI is not a wrapper around ChatGPT. It is not a prompt engineering tool. It is a complete cognitive architecture built from scratch, where every component has a specific purpose, every algorithm has documented complexity, and every claim is backed by a test."
"The question I wanted to answer was simple: can a machine understand language without a billion-parameter language model? The answer turned out to be yes, if you build the right architecture."
Tobias Østen
Sole Architect, Condrox AI
Story Angles for Editors
Resources for Deeper Coverage
| Resource | What It Contains | Best For |
|---|---|---|
| Overview | Founder bio, system summary, proof sections with real test functions and code excerpts | Quick introduction |
| Architecture | Full system architecture: 18-stage pipeline, 26 CI components, 5 agents, AI-OS, knowledge system | Technical overview |
| Deep Dive | 24 sections of actual source code with documented O() complexity | Engineers verifying every algorithm |
| Certification | 20-point verification checklist, test inventory, code metrics, source code review request form | Independent verification |
| Explained | Plain-language explanation of how the system works, for non-technical readers | General audience |
| Benchmarks | Detailed performance measurements: latency, throughput, memory, comparison data | Performance claims |
| Comparison | Side-by-side comparison with ChatGPT, Claude, Gemini, and Llama | Competitive analysis |
| White Paper | Formal technical article suitable for publication or peer review | Academic or deep technical coverage |
| Technical FAQ | Answers to questions a reviewer or editor would ask | Interview preparation |
Contact
Tobias Østen
Sole Architect, Condrox AI
Email: kontakt@pcnorge.no
Location: Norway
Response time: Within 24 hours for press inquiries
Available for: Interviews, technical reviews, demos, and Q&A sessions
Usage Rights for Press
Editorial License
Editors and journalists may quote from this documentation, reproduce tables and statistics, and reference the technical claims made here, provided that:
- Condrox AI is attributed as the source
- Tobias Østen is credited as the architect
- A link to this documentation is included where applicable
- Statistics are cited as "as of Version 3.4.0, July 2026"
For any other usage, including reproducing source code or using the Condrox AI name in commercial contexts, contact kontakt@pcnorge.no.