Intelligence as Pattern Geometry
SYNTHOS is a syntax-driven AI architecture where cognition flows through a 7-layer regex pipeline — not neural weights. Deep research, long-form text generation, command execution, encryption, and a beautiful TUI — all in one install.
Capabilities
Everything you need, encoded in pattern geometry
From encryption to model evaluation, every capability is built on the same regex-topology foundation.
Symbolic Topology Cipher
256-bit block cipher built from SYNTHOS primitives: lattice permutation, regex S-box, Möbius fold. 12 rounds, HMAC-SHA256 auth.
Model Evaluation Engine
Benchmark any Ollama / vLLM / llama.cpp endpoint. Composite scoring across quality, speed, coherence, and cost.
Rich TUI & CLI
System-wide `synthos` command. Tabbed Textual TUI with chat, status, eval, crypto panels. Headless REPL fallback.
Syntax-Based LLM
SynthoLM — a regex-topology inference engine. Processes prompts through 7 layers and generates structured responses without neural weights.
Architecture
7-Layer Cognitive Pipeline
Every input traverses all seven layers. Each layer is a standalone Python module with its own regex geometry, data structures, and visualization.
60+ regex primitives with ASCII geometric forms — the atomic alphabet of cognition.
4×4 cell grid with directional regex transitions. Input is routed spatially, not sequentially.
Named captures become concept nodes; 7 relation types weave a meaning-space graph.
8-head attention via pattern intersection geometry. Overlap area = relevance.
EBNF productions encoded in recursive regex with FSM parsing and parse-tree construction.
Rank-3 symbolic tensor + episodic / short-term / long-term memory lattice.
Substitution chains transform crystallized state into structured ASCII output.
Interactive
Live Demo
Talk to SYNTHOS directly. Every message flows through all 7 layers in real time.
SYNTHOS v3.0 — 7-layer pipeline online. Type a message or try an example below.
Get Started
One command. Any platform.
Install synthos v3 with a single line — works on macOS, Linux, and WSL.
curl -fsSL https://synthos.dev/install.sh | bashThen run
synthosLaunch TUIsynthos chatHeadless REPLsynthos encrypt "hello"STC ciphersynthos evalBenchmark modelsRequires Python ≥ 3.9 and git. The installer auto-detects your platform, creates a virtual environment, and puts synthos on your PATH.