tylergarlick/abraxas/research-engine

Research Engine Skill

The Research Engine provides the tools necessary to expand the system's knowledge base through active information gathering and synthesis. It automates the process of discovering, fetching, and distilling information from the open web.

Core Capabilities

The engine moves beyond simple search by implementing iterative discovery and structured synthesis.

Web Search and Retrieval

The engine utilizes a multi-stage search process:

1. Direct Search: Uses specialized APIs to retrieve the most relevant snippets and URLs.
2. Targeted Fetching: Extracts full-page content, removing HTML noise and focusing on the primary textual body.
3. Link Extraction: Identifies a network of related resources for further deep-diving.

Content Synthesis

The synthesize tool transforms disparate pieces of information into structured reports.

• Mechanism: Aggregates key points from multiple sources, cross-references them, and formats the output into an Executive, Technical, or Comprehensive report.
• Verification: Identifies failures in source retrieval and highlights gaps in the synthesized data.

Iterative Deep-Dive Research

A multi-pass process that evolves from a broad question to specific, verified claims.

• Iteration 1: Broad landscape mapping via initial search.
• Iteration 2: Following specialized leads and focus areas identified in Iteration 1.
• Iteration 3: Verification phase—cross-referencing key claims across the discovered sources.

Commands

web_search

Search the web for current information.

• Arguments: query (required), max_results (optional).

web_fetch

Extract content from a specific URL.

• Arguments: url (required), extract_links (optional).

synthesize_report

Synthesize a report from multiple sources.

• Arguments: topic (required), sources (list of {type, value}), report_format (optional).

deep_dive_research

Execute a multi-iteration research workflow.

• Arguments: research_question (required), max_iterations (optional), require_verification (optional), focus_areas (optional).

health_check

Diagnostic status of the engine and its API dependencies.

• Arguments: verbose (optional).

Implementation Details

• Architecture: Bun + TypeScript Abraxas MCP server (mcp/src/tools/research .ts).
• Dependencies: Uses built-in fetch for HTTP communication.

MCP tools

web_search, web_fetch, synthesize_report, deep_dive_research, research_engine_health_check.

• Connectivity: Interfaces with Ollama's search API and Brave Search.