ain-research 0.2.1

AIN Infinite Research System - Multi-Agent Automated Science DAEMON. Engineered to sit on top of LLMs to autonomously guide research, provide storage, and retrieval functionality. Evolving into a 10x autonomous ecosystem.

AIN Research Daemon Library

Welcome to the AIN Infinite Research Daemon (ain-research), an open-source Python library built by That-Tech-Geek. This library provides the core infrastructure for an Autonomous Intelligence Network (AIN), sitting seamlessly on top of LLMs to guide, automate, and orchestrate open-ended domain research.

What is it?

ain-research is a sophisticated pipeline for automated research ingestion, storage, and retrieval. It allows LLMs and autonomous agents to asynchronously fetch papers from ArXiv, crawl open-source repositories from GitHub, and store them securely into a local SQLite queue before finally syncing them into a hyper-fast, modular knowledge base (the Vault).

Key capabilities include:

• Infinite Research Daemon: Time-boxed, autonomous daemon that fetches domain-specific research using intelligent rate-limiting and rotating APIs.
• LLM Steering Integration: It is engineered to sit on top of LLMs. You can use the CLI or built-in MCP Server to dynamically add new search queries, refine categorizations, and prioritize processing.
• Atomic Concurrency Control: Uses OS-level kernel FileLocks to ensure zero data corruption during heavy concurrent reading/writing by different agents.
• Model Context Protocol (MCP): Directly expose your AIN Second Brain to any MCP-compatible LLM client (like Claude Desktop or other agents), allowing them to queue research, check system health, and compile the wiki on demand.

Use Cases

1. Automated Overnight Crawling Let the daemon run overnight (ain-daemon). It crawls ArXiv and GitHub based on your configured categories (e.g., Quant Finance, LLMs) and stores the results securely. By the time you wake up, your Vault is packed with fresh, tagged Markdown files.

2. LLM-Guided Research If you have a primary agent researching "Agentic Frameworks", the agent can use the MCP server (ain-mcp) to dynamically add "Agentic Frameworks" to the daytime priority ingestion queue. The daemon will immediately fetch relevant papers and repos, syncing them into the Vault for the LLM to read.

3. Hyper-fast Retrieval The core indexing engine (ain compile) resolves thousands of nodes into lightweight Maps of Content (MOCs) and exact tag inversions, enabling LLMs to quickly query specific subjects without context-window bloat.

Installation

pip install ain-research

Quick Start

1. Configure Workspace

By default, the package uses ~/.ain/ as the root workspace. You can override this using the AIN_WORKSPACE environment variable.

2. Start the Daemon

ain-daemon

Runs the overnight crawler. By default, it operates actively between 22:00 and 23:59.

3. Add to Queue via CLI

ain queue add --arxiv 2305.14314
ain queue add --github karpathy/nanoGPT

4. Compile the Knowledge Base

ain compile

Generates the Maps of Content (MOCs), Mermaid network graphs, and visualizer data.

5. Launch MCP Server

To allow LLMs to natively hook into the daemon's research capabilities:

ain-mcp

(Configure this in your LLM client's MCP configuration settings).

Author

That-Tech-Geek

10x Evolution Features (v0.2.0+)

ain-research has evolved from an assistive background utility into an autonomous, 10x-scale ecosystem. It incorporates several advanced systems:

1. Self-Verifying Logic Engine

It actively fights hallucinations using a highly deterministic logic system.

• Cross-Examination: It flags anomalies and invalid assumptions by cross-referencing contradictory data points.
• Logic Proofs: Evaluates findings with mathematical and architectural correctness, minimizing the need for human debugging.

2. Multi-Agent Swarm Orchestration

Instead of running as a single synchronous loop, it now operates as a background multi-agent network.

• Division of Labor: Specialized agents concurrently verify, structure, and peer-review incoming research data.
• Continuous Execution: The swarm processes queue items in real-time alongside the daemon's regular horizon scanning.

3. Zero-Friction Abstract Interface (ain intent)

You can bypass manual tool configurations by simply passing high-level intents.

• Command: ain intent "Explore sparse autoencoders in LLMs"
• The system automatically leverages the LLM client to decompose this intent into discrete tasks, builds the search queries, and delegates them to the swarm.

4. Traceable Truth Layer

Decisions are no longer a "black box".

• Deterministic trace logs are saved into truth_logs/.
• Every LLM judgement, verification step, and data modification is recorded to provide full auditability, turning untrusted guessing into enterprise-grade intelligence.