Deep Research Analyst

Multi-hop web research with citations. Your private Perplexity replacement, fully local.

INCLUDED TOOLS

DOCUMENTATION

---
slug: deep-research-analyst
title: Deep Research Analyst
tagline: Multi-hop web research with citations. Your private Perplexity replacement, fully local.
status: live
tier: S
category:
  - research
  - productivity
version: 1.0.0
releasedAt: '2026-04-16T00:00:00Z'
lastUpdated: '2026-04-16T00:00:00Z'
requires:
  cognithor: '>=0.92.0'
  llm: 'qwen3:7b or qwen3:27b'
  ram: '16 GB'
tools:
  - deep_research_pro
  - research_export_md
  - research_export_pdf
  - research_history
seo:
  title: 'Deep Research Analyst · Cognithor Agent Pack'
  description: 'Multi-hop web research with citations, source triangulation, and PDF export. Your private Perplexity replacement.'
related:
  - reddit-lead-hunter-pro
  - personal-crm-silent
  - content-creator-os
---

## Deep Research Analyst

A paid agent pack that turns Cognithor into a Perplexity-style research engine.
Ask any question — the agent iterates through up to 5 search hops, triangulates
sources, builds a citation graph with confidence scores, and produces a polished
report in Markdown or PDF. Every source tracked, every claim verified, zero cloud.

### How it works

1. **Decompose** — Your question is split into targeted sub-queries by a local LLM.
2. **Search** — Cognithor's web search fetches real-time results from DuckDuckGo (with SearXNG/Brave/Google CSE fallback).
3. **Fetch** — Full page content is retrieved for top results, not just snippets.
4. **Synthesise** — The LLM writes a structured report with inline `[1]` `[2]` citation markers.
5. **Gap-check** — The engine evaluates remaining knowledge gaps and runs another hop if needed.
6. **Triangulate** — Claims are cross-checked across sources. Single-source claims are flagged `[unverified]`. Contradictions get a `⚠` warning.
7. **Export** — Publish to Markdown (with YAML frontmatter) or PDF (with header, body, and full source list).

### Tools

| Tool | Description |
|------|-------------|
| `deep_research_pro` | Multi-hop research — returns a cited Markdown report + triangulation summary |
| `research_export_md` | Export a stored result to `.md` with YAML frontmatter |
| `research_export_pdf` | Export a stored result to `.pdf` via fpdf2 |
| `research_history` | List, get, search, or delete stored research results |

### Example usage

> "Research the current state of local LLM inference performance benchmarks, focusing on Qwen3 and Llama 4."

Cognithor will run 3–5 search hops, cite 8–15 sources with confidence scores, flag any contradicting benchmark numbers, and save the result as a PDF to your desktop in under 2 minutes.

### Privacy

- No cloud API calls. All LLM inference runs locally via Ollama.
- Web searches use Cognithor's existing search stack (DuckDuckGo by default).
- Research history is stored in `~/.cognithor/research/history.db` (SQLite, local only).

### Requirements

- Cognithor >= 0.92.0
- Ollama running locally with `qwen3:7b` (fast) or `qwen3:27b` (better quality)
- 16 GB RAM recommended for qwen3:27b
- `fpdf2` Python package (included in Cognithor core dependencies) for PDF export