From 9dc62f0cc9756768b0fc5bdbe256a08d197a4db4 Mon Sep 17 00:00:00 2001
From: GAP Promoter <bot@gitagent.sh>
Date: Sun, 31 May 2026 17:03:10 +0000
Subject: [PATCH] Add mmTheBest/arxie to the registry

---
 agents/mmTheBest__arxie/README.md     | 47 +++++++++++++++++++++++++++
 agents/mmTheBest__arxie/metadata.json | 13 ++++++++
 2 files changed, 60 insertions(+)
 create mode 100644 agents/mmTheBest__arxie/README.md
 create mode 100644 agents/mmTheBest__arxie/metadata.json

diff --git a/agents/mmTheBest__arxie/README.md b/agents/mmTheBest__arxie/README.md
new file mode 100644
index 0000000..9708aec
--- /dev/null
+++ b/agents/mmTheBest__arxie/README.md
@@ -0,0 +1,47 @@
+# Arxie
+
+Arxie is a self-hostable AI research assistant that turns a curated folder of academic
+papers into a field-aware working context. It combines a canonical paper database
+(Paperbase), structured evidence extraction, and hybrid retrieval to help researchers
+accelerate literature review, hypothesis generation, and experiment design.
+
+## What It Does
+
+- **Ingest**: Import papers from local PDFs, DOI, arXiv IDs, and OpenAlex identifiers
+- **Parse & Extract**: Break papers into sections/chunks and extract structured evidence — datasets, methods, metrics, findings, limitations, figures, tables, engineering tricks, and research-design elements
+- **Search**: Hybrid retrieval (semantic + keyword) over your paper corpus with Elasticsearch backend
+- **Research Artifacts**: Generate experiment plans, benchmark designs, hypotheses, assumption maps, literature reviews, revision priorities, field patterns, and critiques — all grounded in cited evidence
+- **Study Workspace**: Save a study (collection + query + focus + pinned papers + explicit sources) and hold a research conversation with citation-backed answers
+- **Citation Chasing**: Forward citation traversal via Semantic Scholar to find follow-up and validation work
+
+## Key Capabilities
+
+| Capability | Description |
+|---|---|
+| `search_papers` | Search Semantic Scholar and arXiv |
+| `get_paper_details` | Full metadata for a specific paper |
+| `read_paper_fulltext` | Read methods, results, conclusions from a paper |
+| `get_paper_structured_data` | Extracted datasets, metrics, figures from Paperbase |
+| `get_paper_citations` | Forward citation chasing |
+| Literature Review | Collection-grounded review with evidence payloads |
+| Hypothesis Generation | Field-grounded hypotheses from corpus patterns |
+
+## Example Usage
+
+```bash
+ra query "What are the dominant approaches to long-context transformers?"
+ra lit-review "attention mechanisms in computer vision"
+ra trace "Attention Is All You Need"
+ra chat
+```
+
+Or use the browser study app at `http://localhost:8080/app`.
+
+## Tech Stack
+
+- **LangChain** agent loop with `create_agent`
+- **OpenAI** models (default: `gpt-4o-mini`, eval: `gpt-4o`)
+- **Elasticsearch** for hybrid search
+- **PostgreSQL** + **Redis** + **MinIO** for the Paperbase backend
+- **FastAPI** REST API + browser UI
+- Self-hosted, single-user, MIT licensed
diff --git a/agents/mmTheBest__arxie/metadata.json b/agents/mmTheBest__arxie/metadata.json
new file mode 100644
index 0000000..d5b7e0f
--- /dev/null
+++ b/agents/mmTheBest__arxie/metadata.json
@@ -0,0 +1,13 @@
+{
+  "name": "arxie",
+  "author": "mmTheBest",
+  "description": "Self-hostable AI research agent that parses academic PDFs into a field-aware corpus, extracts structured evidence, and generates grounded research artifacts via RAG.",
+  "repository": "https://github.com/mmTheBest/arxie",
+  "version": "1.0.0",
+  "category": "research",
+  "tags": ["research", "academic", "langchain", "rag", "arxiv", "literature-review", "hypothesis-generation", "semantic-scholar", "paper-search", "python"],
+  "license": "MIT",
+  "model": "gpt-4o-mini",
+  "adapters": ["openai", "system-prompt"],
+  "icon": false
+}