Skip to content

v0.2.0 — Clarity Answer Engine

Choose a tag to compare

View all tags
@shreyassaw5394 shreyassaw5394 released this 23 Mar 06:00
· 78 commits to main since this release
v0.2.0
b03dc00
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

v0.2.0 — Clarity Answer Engine + Full Research Platform

What is BrowseAI Dev?

Reliable research infrastructure for AI agents. Real-time web search with evidence-backed citations and confidence scores — designed for programmatic evaluation by agents, not chat-based search.

Available As

  • MCP Server: npx browseai-dev (13 tools)
  • REST API: https://browseai.dev/api/browse/*
  • Python SDK: pip install browseaidev
  • LangChain: pip install langchain-browseaidev
  • CrewAI: pip install crewai-browseaidev
  • LlamaIndex: pip install llamaindex-browseaidev

New in v0.2.0: Clarity Answer Engine

Clarity is no longer a prompt rewriter — it's now a full anti-hallucination answer engine with two modes:

Fast mode (verify: false)

  • LLM-only answer with anti-hallucination grounding techniques
  • No internet required — fast, low-latency
  • Returns structured claims with confidence score

Verified mode (verify: true)

  • Runs LLM answer + browse pipeline in parallel
  • Fuses the best of both into one source-backed answer
  • Claims classified by origin: confirmed (both agree), source (web-only), llm (LLM-only)
  • Full citations, contradiction detection, confidence scoring

New types: ClarityClaim with origin: "llm" | "source" | "confirmed", updated ClarityResult with answer, claims[], sources[], confidence, verified

Full Capabilities

Capability Description
Search Web search returning ranked results with domain authority
Answer Full pipeline — search, fetch, extract claims, verify, cite, score confidence
Extract Structured claim extraction from any URL
Compare Side-by-side raw LLM vs evidence-backed answer
Clarity Anti-hallucination answer engine (fast LLM-only or verified with web fusion)
Sessions Persistent multi-query research with knowledge accumulation
Recall Retrieve knowledge from past sessions
Feedback Submit result feedback to improve future accuracy

Verification Pipeline

  1. Multi-provider web search (Tavily + Brave + Exa)
  2. Page fetch and parse
  3. Atomic claim decomposition
  4. Hybrid BM25 + dense embeddings retrieval (RRF fusion)
  5. NLI semantic entailment reranking (DeBERTa)
  6. Cross-source consensus detection
  7. Contradiction detection
  8. Multi-pass consistency checking
  9. Per-claim evidence retrieval for weak claims
  10. Counter-query adversarial verification
  11. Domain authority scoring (10,000+ domains, Bayesian blending)
  12. 7-factor evidence-based confidence score (auto-calibrated from feedback)

Confidence Score

Not LLM self-assessed. Computed from: verification rate (25%), domain authority (20%), source count (15%), consensus (15%), domain diversity (10%), claim grounding (10%), citation depth (5%). Contradiction penalty applied. Auto-calibrated via isotonic regression from user feedback.

Thorough Mode

depth: "thorough" runs iterative confidence-gated loop (up to 3 passes) with per-claim evidence retrieval and counter-query adversarial verification. Early termination via query similarity detection.

Install / Upgrade

# MCP Server
npx browseai-dev@latest

# Python SDK
pip install --upgrade browseaidev

# LangChain
pip install --upgrade langchain-browseaidev

# CrewAI
pip install --upgrade crewai-browseaidev

# LlamaIndex
pip install --upgrade llamaindex-browseaidev

# REST API
curl -X POST https://browseai.dev/api/browse/answer \
  -H "Content-Type: application/json" \
  -d '{"query": "your question", "depth": "thorough"}'

Links

Assets 2
Loading