ZenthraLabs Intelligence · v0.1.0Apache 2.0 Open Source

ZENE

Autonomous Coding Agent Governed by a Biological Connectome

A native multi-turn autonomous coding agent in Rust. Tree-sitter AST perception, 13 core filesystem and terminal tools, 3-tier skill architecture, and an embedded sub-15µs biological Drosophila connectome that executes involuntary spinal reflex safety gates.

< 15 µs
Reflex Gate
500 Neurons
3,889 Synapses
13 Tools
Tree-sitter AST
0 ms
Runtime Daemon
zene repl — gemini-3.5-flash-lite
100Hz Reflex
> zene "Refactor AST parser and run unit tests"
[Connectome Reflex]: State=EXECUTION · 30 Euler steps in 11.4 µs
⚡ Tool Masking: 13 tools provisioned (AST, diff, write_file)
📁 Perception: Scanned 42 source files via Tree-sitter AST
🔧 Action: edit_file(src/parser.rs:L142-180)
🧪 Action: run_terminal("cargo test --lib parser")
✓ All 18 unit tests passed · 0 compile errors · Turn completed in 1.4s
zene>
Physical Neuroscience Architecture

The Fly Brain Sensory Reflex Gate

Traditional agents send full tool schemas to cloud LLMs on every single turn, leaving files vulnerable to hallucinated wipe commands and burning API tokens. ZENE runs an embedded 500-neuron connectome locally on CPU in under 15 microseconds before anything reaches the remote LLM.

01

Giant Fiber Reflex Lock

Modeled after the fruit fly's escape circuit that triggers in milliseconds when a shadow descends. Intercepts destructive commands (rm -rf /, unconstrained drive wipes) and locks workspace mutation before the prompt leaves the device.

02

Dynamic Tool Masking

Connectome neural attractors classify intent into 5 firing states (DISCUSSION, INSPECTION, PLANNING, EXECUTION, REFLEX_LOCK). Injects only minimal relevant tool definitions, cutting prompt token bloat by up to 74%.

03

Pure CPU L1 Residency

The 500-neuron connectome and 3,889 synapses are stored as an embedded 154 KB binary. Executed via 30 Euler numerical integration steps entirely in CPU cache with zero ONNX, LibTorch, or background daemon dependencies.

Published Research Paper · Stage 01

Biologically Grounded Decision Connectomes for Autonomous Coding Systems

Empirical data on Dale's principle adherence, Piéron's law response curves, and real-time attractor phase portraits.

Read Full Technical Report →
Engine Capabilities

13 Native Tools Implemented in Pure Rust

Every tool is written natively with zero shell wrappers or Node scripts, featuring precise line slicing, diff verification, and AST diagnostics.

ToolCategorySafety TypeFunctionality
read_fileFilesystemRead-onlyReads file contents with precise line slice notation
write_fileFilesystemDestructiveWrites or creates new files on the filesystem
edit_fileFilesystemDestructiveTargeted substring search and surgical replacement
searchSearchRead-onlyHigh-speed ripgrep-style content and filename pattern matcher
run_terminalTerminalRequires ApprovalExecutes shell commands in the target workspace
get_diagnosticsDiagnosticsRead-onlyCaptures live compiler, linter, and typecheck diagnostics
git_diffGitRead-onlyInspects staged and unstaged git changes
create_planPlanningRead-onlyInitializes multi-step architecture refactor roadmaps
Developer Setup

Quickstart & CLI Installation

Install the binary locally with Cargo or embed ZENE into your own Rust developer tools.

1. Build and install globallyCargo
# Clone the repositorygit clone git@github.com:kabirajpan/zene.git cd zene# Installs `zene` binary to ~/.cargo/bincargo install --path .# Configure free Gemini or Groq keyexport GEMINI_API_KEY="your-api-key"# Start interactive REPLzene
2. Embed as a Rust LibraryCargo.toml
[dependencies]agent = { git = "https://github.com/kabirajpan/zene.git" }// In your Rust application:use agent::create_agent_for_model_and_workspace; let mut agent = create_agent_for_model_and_workspace( "gemini", "gemini-3.5-flash-lite", "./", )?;
Native IDE Intelligence

Experience ZENE inside Zenthree IDE

ZENE is compiled directly into Zenthree as an interactive sidecar assistant with native PTY terminal bindings, live diagnostics, and zero electron bloat.