Aurum Core

The full agentic research engine. Open source. BYOK.

$ git clone https://github.com/aurolabs-ai/aurum.git $ cd aurum && npm install $ export OPENAI_API_KEY=sk-... $ node bin/aurum.js "your question"
Star on GitHub
aurum
aurum v0.9.0

  Q: What happens if the EU bans diesel by 2030?

  Understanding question...done 0.8s
  Research round 1 30 queriesdone 3.8m
  Data points: 1420.4s
  Research round 2 15 targeteddone 2.1m
  Data points: 1980.3s
  Writing 12 sectionsdone 4.2m
  Director review 5 dimensions92/100 1.4m

  Bottom line: A 2030 diesel ban is likely (75-85%).
  Confidence: probable (Sherman Kent)

  Saved: rapport.md
  8,200 words · 52 sources · 18 min · EUR 5.40
How it works

Eight steps. Zero shortcuts.

Every report runs through the full agentic pipeline. The agent plans, searches reactively, writes, reviews its own work, and delivers a brief with confidence.

01

Understand

Parses your question, identifies domain, selects template, generates targeted search queries.

02

Search

Executes queries in parallel. Extracts data points with source URLs. Tags every finding with A/B/C source tier.

03

Analyze

Maps evidence against planned sections. Identifies coverage gaps, weak areas, missing perspectives.

04

Repeat

Generates targeted queries for gaps. Seeks contradicting evidence. 5-20 rounds, reactive to what it finds.

05

Write

Section-by-section writing using only verified evidence. Every claim traced to its source.

06

Review

Self-review catches unsourced claims, vague language, repetition. Issues fixed inline.

07

Director

Adversarial reviewer scores the report on 5 dimensions. Weak sections get new research and rewrite.

08

Brief

One sentence that changes your decision. Confidence on the Sherman Kent scale. The report is the proof.

What you get

Research-grade output. API-level cost.

50+ verified sources

Every claim traced

No hallucinated references. Every data point links back to its origin. Source URLs verified before the report is saved.

5-20 search rounds

Agentic search

Not one blind search pass. The agent searches reactively -- filling gaps, seeking contradictions, pursuing leads it discovers.

A / B / C grading

Source grading

Institutional sources (government, academic, wire services) ranked above blogs and SEO content. You see the grade on every source.

5-dimension scoring

Director quality review

An adversarial reviewer scores the report on sources, depth, actionability, rigor, and information density. Threshold: 85+.

Sherman Kent scale

Bottom line + confidence

One sentence that answers your question. Confidence scored on the intelligence community's probability scale. The 10,000 words are the proof.

~EUR 5-8 per report

Your API costs

You pay OpenAI directly. No subscription. No markup. Most cost is web search calls plus writing tokens. Your keys, your machine.

Templates

Three research templates. Ready to go.

Each template structures the agent's search strategy, section layout, and source priorities for a specific type of research.

Scenario Analysis

Explores future scenarios with probability assessments. Maps drivers, wildcards, and second-order effects across multiple timelines.

--template scenario-analysis

Market Research

Market sizing, competitive landscape, regulatory environment, and trend analysis. Institutional data sources prioritized.

--template market-research

Competitor Audit

Deep competitive analysis: product positioning, pricing, funding, team, technology stack, market share, and strategic trajectory.

--template competitor-audit

Need more depth?

Aurum Depth uses deep reasoning models with 170+ sources. Full Director loop, extended search rounds, and multi-model verification. EUR 499 per report via Pro.

Try Pro