SRT1-CORE 2.2.3

SRT-1 — Cognitive Operating System for Software Repositories. Anti-hallucination guardrails for AI code assistants.

SRT-1 CORE — Open Source Edition

Brain Over the Repo — Anti-hallucination, architectural coherence, and cryptographic guardrails for AI-assisted software work.

---

What Is SRT-1?

SRT-1 solves the problem of AI code assistants hallucinating architecture or breaking dependencies. It locally scans your codebase, builds a deep Abstract Syntax Tree (AST) manifest, and uses that context to forcefully guide the AI.

The srt1-core package provides the Memory Minimum Architecture—a completely decoupled, local-first engine that runs entirely on your machine via SQLite, without pinging a cloud API.

The "Drop-In" Sandbox (Zero Global State)

Unlike massive IDE extensions or global CLI tools that muddy your entire system, SRT-1 is hermetically sealed. You simply drop the SRT1-CORE-OSS folder into any codebase, and tell your existing AI assistant (Cursor, Copilot, etc.) to run it.

• Auto-Isolated Ports: The engine dynamically finds available network ports, meaning you can drop SRT-1 into 5 different projects simultaneously without a single collision.
• Local Context Only: The engine builds its .srt1 memory cache purely inside the project boundary. Zero cross-contamination across your machine.
• The AI Handcuffs: You aren't replacing your AI; you are dropping a constraint engine into your folder. Tell Cursor to use SRT-1, and the AI is instantly handcuffed to your deterministic, architectural rules.

The Cryptographic Trust Layer

Every action in the SRT-1 pipeline is cryptographically tied to the codebase using Seed Signature. From the moment the AST generates the code manifest, to the exact moment you approve a task dispatch, the engine signs and verifies the payload. This ensures an indisputable, immutable execution roadmap.

The Local Developer Experience

SRT-1 ships with a fully-featured local Developer Dashboard and a Mobile-reflective PWA.

The Human-in-the-Loop Pipeline

SRT-1 acts as a strict proxy between you and your code assistant (like Claude Code, Cursor, or Aider).

1. Plant a Seed: You submit a task via the terminal or the PWA.
2. Review Blueprint: The system generates a blueprint of the intended changes.
3. Approve via Seed Signature: You review the blueprint in the PWA. Once you click Approve, the payload is signed via Seed Signature and dispatched to your code assistant to execute.
4. Self-Heal: If the AI makes a mistake, the AST detects the architectural drift and injects correction warnings into the AI's context files. The next time the AI reads its instructions, it sees the error.

Real-Time Delivery: The MCP Server

Writing corrections to a file is useless if the AI never re-reads it. That's why SRT-1 ships with a Model Context Protocol (MCP) server — a live, bidirectional pipe between SRT-1 and your AI agent.

When connected via MCP, SRT-1 doesn't wait for the AI to check a file. It pushes codebase intelligence directly into the AI's context on every interaction:

• srt1_get_context — AI calls this before making changes. Gets the full code map, risk tags, and warnings.
• srt1_log_interaction — AI calls this after every action. Every 3 calls, SRT-1 fires a reflection checkpoint and pushes a coherence score + correction directives back into the conversation.
• srt1_check_function — AI calls this before creating a function. SRT-1 tells it if the function already exists.
• srt1_set_task — Plants the seed. Everything after this is measured for drift.

Setup for Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "srt1": {
      "command": "srt1-mcp",
      "env": { "SRT1_REPO_PATH": "/path/to/your/project" }
    }
  }
}

Setup for Cursor

Add to .cursor/mcp.json in your project:

{
  "mcpServers": {
    "srt1": {
      "command": "srt1-mcp",
      "env": { "SRT1_REPO_PATH": "." }
    }
  }
}

Once connected, the AI is forced to call SRT-1 tools — and SRT-1 injects corrections directly into the conversation in real time. No file polling. No hoping the AI re-reads.

Installation

pip install srt1-core

Quick Start

1. Index a Repository

srt1-index --repo_path /path/to/your/repo

This generates srt1_code_manifest.json—the foundational map of your codebase, verified by Seed Signature.

2. Start the Live Middleware & Dashboard

srt1-middleware --repo_path ./my_project --port 7483

This spins up the local execution engine and the Developer Dashboard/PWA.

3. Open the PWA

Navigate to http://localhost:7483/mobile to start planting seeds and governing your AI's execution pipeline.

The Unified Ecosystem

pip install srt1-core provides the complete Open Source ecosystem:

Component	Description	CLI Tools
Core Intelligence	Local AST mapping, File hashing, Curation	srt1-index
Pro Execution	Context Bundler, Execution Engine, Self-Healing	srt1-bundle, srt1-execute, srt1-heal
Platform & UI	Live Middleware, Developer Dashboard, PWA	srt1-middleware
MCP Server	Real-time AI injection via Model Context Protocol	srt1-mcp

Architecture

SRT-1 is deployed as a single unified wheel containing all subsystems:

srt1_code_indexer/          → Code reflection and indexing engine
srt1_pro/                   → Bundling, execution, self-heal
srt1_platform/              → Middleware, Seed Queue, MCP server, bridge
srt1_platform/pwa/          → Local Developer UI & Human-in-the-Loop PWA

Requirements

• Python 3.9+
• Core indexing uses the Python standard library only.

License

Apache License 2.0 Author: William Darnell Jernigan IV AKA THE ORIGINAL SEED