From 0f6ae19f6ab94858486f42002b5bc8a5ec41eeb0 Mon Sep 17 00:00:00 2001
From: GAP Promoter <bot@gitagent.sh>
Date: Wed, 27 May 2026 05:35:37 +0000
Subject: [PATCH] Add tegridydev/abstract-agent to the registry

---
 agents/tegridydev__abstract-agent/README.md   | 66 +++++++++++++++++++
 .../tegridydev__abstract-agent/metadata.json  | 14 ++++
 2 files changed, 80 insertions(+)
 create mode 100644 agents/tegridydev__abstract-agent/README.md
 create mode 100644 agents/tegridydev__abstract-agent/metadata.json

diff --git a/agents/tegridydev__abstract-agent/README.md b/agents/tegridydev__abstract-agent/README.md
new file mode 100644
index 0000000..c5d5c79
--- /dev/null
+++ b/agents/tegridydev__abstract-agent/README.md
@@ -0,0 +1,66 @@
+# abstract-agent
+
+A **fully local, 100% private** multi-agent research system for generating novel
+academic hypotheses and publication-ready abstracts. No API keys. No cloud.
+Just you, your GPU/CPU, and Ollama.
+
+## What it does
+
+Given a research topic, abstract-agent orchestrates a five-stage agent pipeline
+that searches real literature databases, critiques existing work, synthesises new
+directions, generates a bold hypothesis, polishes it into a publication-ready
+abstract, and scores its novelty — all on your own hardware.
+
+## Key capabilities
+
+- **Multi-source literature search** — pulls from arXiv, Semantic Scholar,
+  PubMed, bioRxiv, medRxiv, EuropePMC, Crossref, DOAJ, and OpenAlex
+- **Five specialised sub-agents:**
+  - **Agent A · Breakdown** — decomposes the topic with encyclopedic thoroughness
+  - **Agent B · Critique** — identifies gaps, contradictions, and over-explored areas
+  - **Agent C · Synthesis** — integrates feedback into a refined research direction
+  - **Agent D · Novelty Generation** — proposes one bold, unconventional hypothesis
+  - **Agent E · Academic Structuring** — produces a single-paragraph publication-ready abstract
+- **Novelty Assessment** — scores the hypothesis (1–10) against retrieved literature
+- **Run stats** — reports papers searched/used, total time taken
+- **Config-driven** — swap models, change personas, extend the pipeline via `agents_config.yaml`
+- **Zero data egress** — all LLM calls go to your local Ollama instance
+
+## Example output
+
+```
+Novelty Score: 7/10
+
+Hypothesis: This research introduces a novel approach to LLM compression 
+predicated on Neuro-Symbolic Contextual Compression, translating attention 
+maps into a discrete graph representation and employing learned graph pruning 
+to preserve critical semantic relationships while reducing model size.
+```
+
+## Quickstart
+
+```bash
+git clone https://github.com/tegridydev/abstract-agent
+cd abstract-agent
+pip install -r requirements.txt
+ollama pull gemma3:4b
+python agent/agent.py
+```
+
+## Configuration
+
+Edit `agent/agents_config.yaml` to change:
+- Ollama model (default: `gemma3:4b`)
+- Agent personas and prompt templates
+- Number of literature sources fetched
+- Pipeline steps and output fields
+
+## Requirements
+
+- Python 3.9+
+- [Ollama](https://ollama.com/download) running locally
+- Internet access for public literature APIs (read-only)
+
+## License
+
+MIT — built by [tegridydev](https://github.com/tegridydev). Fork it, break it, share it.
diff --git a/agents/tegridydev__abstract-agent/metadata.json b/agents/tegridydev__abstract-agent/metadata.json
new file mode 100644
index 0000000..37eb543
--- /dev/null
+++ b/agents/tegridydev__abstract-agent/metadata.json
@@ -0,0 +1,14 @@
+{
+  "name": "abstract-agent",
+  "author": "tegridydev",
+  "description": "Fully local multi-agent pipeline for generating novel research hypotheses and publication-ready abstracts using Ollama LLMs — no API keys, no cloud.",
+  "repository": "https://github.com/shreyas-lyzr/abstract-agent",
+  "version": "2.0.0",
+  "category": "research",
+  "tags": ["research", "hypothesis-generation", "ollama", "local-llm", "multi-agent", "arxiv", "pubmed", "academic", "literature-search", "privacy"],
+  "license": "MIT",
+  "model": "ollama:gemma3:4b",
+  "adapters": ["system-prompt"],
+  "icon": false,
+  "banner": false
+}