Change language
Get Started

AI-Powered Research Agent

Where agents
find answers.

Parallel is an autonomous research agent that plans investigations, fans out sub-agents, extracts and verifies contacts, and writes structured reports — turning hours of manual research into minutes.

Rust · 49 tools · 9 crates · one binary

parallel-research run
$parallel run "Find VP Engineering at Series B startups in SF"
PlanningDecomposing into 4 sub-tasks…
ResearchingSpawning 4 parallel agents
Searching LinkedIn & Crunchbase
Parsing corporate websites
Extracting social profiles
Verifying email addresses
CompleteFound 47 contacts, 12 verified
SavingExporting to CSV + CRM sync
$

Search backends

What you get

Three superpowers, one agent

From a query to verified, enriched contacts and personalized outreach — without touching a spreadsheet.

Agent types

Purpose-built agent roles

Each sub-agent has a focused role and its own behavior. Coordinators decompose, fan out, reflect, and synthesize.

Researcherbreathing · scanning sources
Extractingpulse · pulling fields
Structuringfiring · building schema
Searchingaurora · 7 backends
Cleaningsweep · dedup pass
Memoringtrail · absorbing facts
Outreachin flight · composing

Performance

Built for speed.
Measured, not claimed.

Real benchmarks from the release build — parallel tool execution spreads CPU across cores and ships verified results with minimal dispatch overhead.

See all benchmarks →
3.78×
CPU-bound parallel speedup
531k
rows/s HTML parsing
1.1M
items/s feed parsing
~2ms
hybrid memory search

The full toolbox

One engine, every capability

44+ tools, 7 backends

Web, files, exec, browser, vision, git, PDF, OSINT, memory.

Semantic memory

Vector + BM25 hybrid search, entity graph, auto-absorb of findings.

MCP

Client (stdio/HTTP/OAuth) + server mcp-serve.

CRM sync

amoCRM, Bitrix24, HubSpot with dedup.

Durable jobs

SQLite-backed background tasks that survive restarts.

HTTP API + dashboard

SSE streaming, steering, Prometheus metrics.

How it works

From question to verified contacts

01

Ask

Type a research goal in plain language. The coordinator decomposes it into sub-tasks.

02

Research

Parallel agents search the web, parse sites, extract contacts, and spawn their own children.

03

Verify

Every email MX-checked, phones validated, social profiles confirmed.

04

Export

CSV, PDF, HTML, JSON, DOCX — or pushed to CRM with zero duplicates.

Testimonials

Trusted by researchers & revenue teams

"Parallel found 200 verified contacts for our DACH expansion in under an hour. What used to take our SDRs a full week."

Sarah Kim
VP of Sales, TechFlow

"The verification layer alone is worth it. Our bounce rate dropped from 23% to 2%."

Marco Rossi
Head of Growth, DataPrime

Start building with Parallel

Distributed individually. No vendor lock-in. Full control over your research pipeline.

curl -sSf https://parallel.research/install.sh | sh
Request access →

parallel-research bench · docs · request access