Skip to content

Loopshape/_NIX-SYSOP-AI-AGENT

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

146 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🤖 AI Autonomic Synthesis Platform v32 Code Review ​This document reviews the provided single-file AGI platform script, which impressively integrates a robust Bash agent core with a Node.js web server frontend. ​Architectural Summary ​The platform operates in two distinct modes, unified within a single file: ​Web Server Mode (if [[ "${1:-}" == "serve" ...): This is the entry point for the interactive user interface. It runs a lightweight Node.js HTTP server, serving a minimal HTML/CSS/JS frontend. Crucially, it handles API requests (/api/command) by using Node's child_process.exec to call the Bash script itself with the --prompt argument. ​Bash AGI Core (main): This is the high-level command dispatcher and the core of the autonomous workflow. It initializes the environment, manages memory (sqlite3), and runs the run_agi_workflow which implements a multi-model (Messenger, Planner, Executor) loop to achieve the user's goal. ​Key Strengths ​The design demonstrates several sophisticated features typically found in dedicated orchestration frameworks: ​Autonomous Loop (run_agi_workflow): This multi-agent pattern (Messenger -> Multi-Planner -> Executor) is a very effective way to improve reasoning and reduce hallucination by forcing consensus or synthesis. ​Fast & Streaming LLM Workers: The inclusion of both run_worker_fast (for Planner/internal calls) and run_worker_streaming (for Executor output/user feedback) is excellent for providing both low-latency planning and real-time user experience. ​Robust State Management: ​Persistent Memory: Using SQLite for fuzzy caching (memories) provides a crucial long-term memory mechanism for efficiency. ​Secure Execution: The calculate_hmac and subsequent HMAC verification check before tool execution is a vital security feature, ensuring the LLM is not executing arbitrary, unverified code paths. ​DevOps Tooling: The built-in commands (--build, --rebuild, --scan, --git) make the script a powerful project management utility, not just an AI agent. ​Areas for Refinement and Future Development ​While highly functional, the integration presents a few common challenges associated with the Node/Bash bridge and concurrent processing: ​1. Concurrency and Data Contention ​Problem: The Node.js server can handle multiple simultaneous requests, meaning multiple instances of the Bash AGI core (run_agi_workflow) might be running in parallel. ​Risk: Concurrent access to the shared resources, especially the SQLite database ($CORE_DB) and temporary files ($SWAP_DIR), can lead to database locking, corruption, or inconsistent state. ​Suggestion: Implement a basic file-based mutex/lock around the critical sections of the DB access (e.g., inside get_cached_response, add_to_memory_fast, and the tool logging). ​2. Node.js to Bash Execution Safety

+++

A full-fledged local AI orchestration framework in one Bash script! 😎

What This Script Does

  1. Environment Setup

Sets up directories for tasks, projects, databases, SSH keys, and session tracking.

Initializes SQLite databases for:

ai_data.db → memories and events.

blobs.db → code artifacts or other content generated.

wallet.db → seed phrases and wallets.

Optionally initializes a Git repo for project version control.

  1. Dependency Checks

Ensures sqlite3, node, python3, git, and ollama are installed before continuing.

  1. Embedded JavaScript Orchestrator (orchestrator.js)

Orchestrates LLM tasks using multiple models: "phi3", "llama3", "codegemma".

Implements recursive consensus:

Runs multiple models in parallel.

Fuses outputs, checks convergence over iterations.

Tracks a "proof of reasoning" state for each task (ProofTracker).

Can parse and save code blocks into projects and commit them via Git.

Logs events and stores results in SQLite databases.

  1. Embedded Python Manager (manager.py)

Lightweight web server for monitoring: serves JSON at port 8000.

Wallet & seed phrase management:

Uses bip_utils if available to generate mnemonic seeds.

Stores seeds in a SQLite database.

Placeholder for wallet creation/listing.

  1. Command Interface (ai)

Acts as a single CLI to:

Run AI tasks (run).

Import external content (import ).

View memories/events/blobs.

Manage SSH keys, seeds, wallets.

Start/stop sessions for projects.

Run a local server.

Show environment status.

  1. Defaults / Fallback

If no arguments are provided, hashes the current timestamp and lists all files recursively.


Notable Features

✅ Multi-LLM Consensus Loop: The orchestrator fuses results from multiple LLMs and tracks convergence, simulating "team reasoning".

✅ Automatic Project Management: Creates project directories, saves generated code, and commits it to Git automatically.

✅ Persistence: Uses SQLite to store past prompts, results, events, and code blobs.

✅ Extensible: You could add more models to MODEL_POOL or extend Python manager functionality.


Potential Gotchas / Notes

  1. ollama dependency: Must be installed and available in PATH. Without it, the orchestrator won’t run.

  2. Python Wallet Features: Currently limited; wallet creation logic is mostly a placeholder.

  3. Concurrency / Scaling: The recursive consensus runs models sequentially per iteration. With many models, it could be slow.

  4. Security: Storing seed phrases and wallets in plain SQLite files could be risky—encrypt if used with real assets.

  5. Git Commit: Automatically commits all changes in $AI_HOME. Could be noisy if $AI_HOME has unrelated files.


How to Use It

  1. Install dependencies:

sudo apt install sqlite3 nodejs python3 git

install ollama via their instructions

pip install bip-utils

  1. Run a task:

./ai run "Generate a Python script that prints Fibonacci numbers" --project=fib

  1. Start the server:

./ai server

  1. Manage wallet seeds:

./ai seed generate ./ai seed list

  1. View events:

./ai events

About

LOOPCORE <web.screen@outlook.com>

Resources

Stars

Watchers

Forks

Packages

No packages published

Contributors 3

  •  
  •  
  •