Parallel-Web-Tools

CLI and data enrichment utilities for the Parallel API.

Note: This package provides the parallel-cli command-line tool and data enrichment utilities in the parallel-web-tools package. It depends on parallel-web, the official Parallel Python SDK, but does not contain it. Install parallel-web separately if you need direct SDK access.

Features

• CLI for Humans & AI Agents - Works interactively or fully via command-line arguments
• Web Search - AI-powered search with domain filtering and date ranges
• Content Extraction - Extract clean markdown from any URL
• Data Enrichment - Enrich CSV, JSON, DuckDB, and BigQuery data with AI
• Follow-up Context - Chain research and enrichment tasks using --previous-interaction-id
• Research Memory - Recall, scope, evict, or clear saved Task, Monitor, and FindAll work
• AI-Assisted Planning - Use natural language to define what data you want
• Multiple Integrations - Polars, DuckDB, Snowflake, BigQuery, Spark

Installation

Requires Python 3.10+.

Standalone CLI (Recommended)

Install the standalone parallel-cli binary for search, extract, enrichment, and deep research (no Python required):

# macOS / Linux (Homebrew)
brew install parallel-web/tap/parallel-cli

# macOS / Linux (shell script)
curl -fsSL https://parallel.ai/install.sh | bash

The shell script automatically detects your platform (macOS/Linux, x64/arm64) and installs to ~/.local/bin.

Note: The standalone binary supports search, extract, research, and enrich run with CLI arguments, CSV files, and JSON files. For YAML config files, interactive planner, DuckDB/BigQuery sources, or deployment commands, use pip install.

npm

npm install -g parallel-web-cli

This downloads the pre-built binary for your platform. No Python or Go required.

Python Package

For programmatic usage or additional features:

# Minimal CLI (search, extract, enrich with CLI args)
pip install parallel-web-tools

# + YAML config files and interactive planner
pip install parallel-web-tools[cli]

# + Data integrations
pip install parallel-web-tools[duckdb]       # DuckDB (includes cli, polars)
pip install parallel-web-tools[bigquery]     # BigQuery (includes cli)
pip install parallel-web-tools[spark]        # Apache Spark

# Full install with all features
pip install parallel-web-tools[all]

CLI Overview

parallel-cli
├── auth                    # Check authentication status
├── login                   # Device OAuth login (single command)
├── logout                  # Remove stored credentials
├── search                  # Web search
├── extract / fetch         # Extract content from URLs
├── research                # Deep research commands
│   ├── run                 # Run deep research on a question or topic
│   ├── status              # Check status of a research task
│   ├── poll                # Poll until completion
│   └── processors          # List available research processors
├── enrich                  # Data enrichment commands
│   ├── run                 # Run enrichment
│   ├── status              # Check status of a task group
│   ├── poll                # Poll until completion and collect results
│   ├── plan                # Create YAML config
│   ├── suggest             # AI suggests output columns
│   └── deploy              # Deploy to cloud systems (requires pip install)
├── findall                 # Web-scale entity discovery
│   ├── run                 # Discover entities matching a natural language objective
│   ├── ingest              # Preview the schema before running
│   ├── entity-search       # Fast entity search (best-effort optimized for low latency)
│   ├── status              # Check status of a FindAll run
│   ├── poll                # Poll until completion
│   ├── result              # Fetch results of a completed run
│   ├── enrich              # Enrich existing FindAll results with new columns
│   ├── extend              # Request additional candidates for a run
│   ├── schema              # Get the schema for a FindAll run
│   └── cancel              # Cancel a running FindAll
├── monitor                 # Continuous web change tracking
│   ├── create              # Create a new web monitor (event_stream or snapshot)
│   ├── list                # List monitors (cursor paginated)
│   ├── get                 # Get monitor details
│   ├── update              # Update frequency, webhook, metadata
│   ├── cancel              # Cancel a monitor (irreversible)
│   ├── events              # List events for a monitor
│   └── trigger             # Trigger an immediate one-off run
├── skills                  # Install, update, and manage Parallel agent skills
│   ├── install             # Install or update managed Parallel skills
│   ├── list                # List available Parallel skills
│   ├── reinstall           # Cleanly reinstall managed Parallel skills
│   └── uninstall           # Remove only CLI-managed Parallel skills
└── memory                  # Saved Task, Monitor, and FindAll entries
    ├── retrieve            # Search Memory or list recent entries
    ├── evict               # Remove one entry from Memory
    └── clear               # Permanently clear selected Memory

Run parallel-cli skills install to install or update agent skills. When using --skill, other managed skills are removed; unmanaged skills are preserved.

Quick Start

1. Authenticate

# Device OAuth login (shows code and opens browser)
parallel-cli login

# Or set environment variable
export PARALLEL_API_KEY=your_api_key

2. Search the Web

# Natural language search
parallel-cli search "What is Anthropic's latest AI model?" --json

# Keyword search with filters
parallel-cli search -q "bitcoin price" --after-date 2026-01-01 --json

# Search specific domains
parallel-cli search "SEC filings for Apple" --include-domains sec.gov --json

3. Extract Content from URLs

# Extract content as markdown
parallel-cli extract https://example.com --json

# Extract with a specific focus
parallel-cli extract https://company.com --objective "Find pricing info" --json

# Get full page content
parallel-cli extract https://example.com --full-content --json

4. Enrich Data

# Let AI suggest what columns to add
parallel-cli enrich suggest "Find the CEO and annual revenue" --json

# Create a config file (interactive)
parallel-cli enrich plan -o config.yaml

# Create a config file (non-interactive, for AI agents)
parallel-cli enrich plan -o config.yaml \
    --source-type csv \
    --source companies.csv \
    --target enriched.csv \
    --source-columns '[{"name": "company", "description": "Company name"}]' \
    --intent "Find the CEO and annual revenue"

# Run enrichment from config
parallel-cli enrich run config.yaml

# Run enrichment directly (no config file needed)
parallel-cli enrich run \
    --source-type csv \
    --source companies.csv \
    --target enriched.csv \
    --source-columns '[{"name": "company", "description": "Company name"}]' \
    --intent "Find the CEO and annual revenue"

# Enrich a JSON file
parallel-cli enrich run \
    --source-type json \
    --source companies.json \
    --target enriched.json \
    --source-columns '[{"name": "company", "description": "Company name"}]' \
    --enriched-columns '[{"name": "ceo", "description": "CEO name"}]'

5. Deploy to Cloud Systems

# Deploy to BigQuery for SQL-native enrichment
parallel-cli enrich deploy --system bigquery --project my-gcp-project

6. Recall Memory Entries

Memory is an explicit recall layer for completed Task, Monitor, and FindAll work. Saving is asynchronous, so a newly completed run may take a short time to appear.

# Retrieve prior work by relevance
parallel-cli memory retrieve --query "serverless inference vendors" --json

# List the most recent saved work
parallel-cli memory retrieve --query "" --json

# Keep application or workspace research in an isolated memory scope
parallel-cli research run "Compare inference vendors" \
    --memory-scope-key workspace_acme
parallel-cli findall run "Serverless inference vendors" \
    --memory-scope-key workspace_acme
parallel-cli monitor create "Track vendor pricing changes" \
    --memory-scope-key workspace_acme

# Retrieve from the same memory scope
parallel-cli memory retrieve \
    --query "inference vendors" \
    --scope-key workspace_acme \
    --json

# Remove one Memory entry without deleting the Task itself
parallel-cli memory evict --kind task --id trun_abc123 --json

# Permanently clear a specific memory scope (underlying resources remain)
parallel-cli memory clear --scope-key workspace_acme --confirm-clear --json

Omit --scope-key / --memory-scope-key to use personal Memory when the credential and account are eligible. Application credentials require a stable scope key containing only letters, digits, underscores, or hyphens.

Non-Interactive Mode (for AI Agents & Scripts)

All commands support --json output and can be fully controlled via CLI arguments.

Key patterns for agents

# Every command supports --json for structured output
parallel-cli search "query" --json
parallel-cli auth --json
parallel-cli research processors --json

# Read input from stdin with "-"
echo "What is the latest funding for Anthropic?" | parallel-cli search - --json
echo "Research question" | parallel-cli research run - --json

# Async: launch then poll separately
parallel-cli research run "question" --no-wait --json   # returns run_id + interaction_id
parallel-cli research status trun_xxx --json             # check status
parallel-cli research poll trun_xxx --json               # wait and get result

# Follow-up: reuse context from a previous task
parallel-cli research run "follow-up question" --previous-interaction-id trun_xxx --json
parallel-cli enrich run --data '[...]' --previous-interaction-id trun_xxx --json

# Exit codes: 0=ok, 2=bad input, 3=auth error, 4=api error, 5=timeout

Follow-up research with context reuse

Tasks return an interaction_id that can be passed as --previous-interaction-id on a subsequent research or enrichment run. The new task inherits the context from the prior one, so follow-up questions can reference earlier results without repeating them.

# Step 1: Run initial research (interaction_id is in the JSON output)
parallel-cli research run "What are the top 3 AI companies?" --json --processor lite-fast
# → { "run_id": "trun_abc", "interaction_id": "trun_abc", ... }

# Step 2: Follow-up research referencing the first task's context
parallel-cli research run "What products does the #1 company make?" \
    --previous-interaction-id trun_abc --json

# Step 3: Use research context for enrichment
parallel-cli enrich run \
    --data '[{"company": "Anthropic"}, {"company": "OpenAI"}]' \
    --target enriched.csv \
    --source-columns '[{"name": "company", "description": "Company name"}]' \
    --enriched-columns '[{"name": "products", "description": "Main products"}]' \
    --previous-interaction-id trun_abc --json

The interaction_id is shown in both human-readable and --json output for research run, research status, and research poll.

More examples

# Search with JSON output
parallel-cli search "query" --json

# Extract with JSON output
parallel-cli extract https://url.com --json

# Suggest columns with JSON output
parallel-cli enrich suggest "Find CEO" --json

# FindAll: discover entities
parallel-cli findall run "AI startups in healthcare" --json

# FindAll entity-search: fast, low-latency ranked entities (synchronous)
parallel-cli findall entity-search "AI startups in healthcare" -t companies -n 25

# Monitor: track web changes
parallel-cli monitor create "Track Tesla SEC filings" --frequency 1d --json

# Recall prior Memory entries
parallel-cli memory retrieve --query "Tesla filings" --json

# Plan without prompts (provide all args)
parallel-cli enrich plan -o config.yaml \
    --source-type csv \
    --source input.csv \
    --target output.csv \
    --source-columns '[{"name": "company", "description": "Company name"}]' \
    --enriched-columns '[{"name": "ceo", "description": "CEO name"}]'

# Or use --intent to let AI determine the columns
parallel-cli enrich plan -o config.yaml \
    --source-type csv \
    --source input.csv \
    --target output.csv \
    --source-columns '[{"name": "company", "description": "Company name"}]' \
    --intent "Find CEO, revenue, and headquarters"

Integrations

Quick Integration Examples

Polars:

import polars as pl
from parallel_web_tools.integrations.polars import parallel_enrich

df = pl.DataFrame({"company": ["Google", "Microsoft"]})
result = parallel_enrich(
    df,
    input_columns={"company_name": "company"},
    output_columns=["CEO name", "Founding year"],
)
print(result.result)

DuckDB:

import duckdb
from parallel_web_tools.integrations.duckdb import enrich_table, findall_table

conn = duckdb.connect()

# Enrich an existing table
conn.execute("CREATE TABLE companies AS SELECT 'Google' as name")
result = enrich_table(
    conn,
    source_table="companies",
    input_columns={"company_name": "name"},
    output_columns=["CEO name", "Founding year"],
)
print(result.result.fetchdf())

# Discover entities with FindAll
result = findall_table(
    conn,
    "countries that have won the FIFA World Cup and their capital cities",
    match_limit=10,
)
result.result.show()

Programmatic Usage

from parallel_web_tools import run_enrichment, run_enrichment_from_dict

# From YAML file
run_enrichment("config.yaml")

# From dictionary
run_enrichment_from_dict({
    "source": "data.csv",
    "target": "enriched.csv",
    "source_type": "csv",
    "source_columns": [{"name": "company", "description": "Company name"}],
    "enriched_columns": [{"name": "ceo", "description": "CEO name"}]
})

Device Authorization (RFC 8628)

For headless environments (SSH, containers, CI), use the device authorization flow:

from parallel_web_tools import request_device_code, poll_device_token

# Step 1: Request a device code
device_info = request_device_code()
print(f"Go to: {device_info.verification_uri_complete}")

# Step 2: Poll until the user authorizes
token = poll_device_token(device_info.device_code)

FindAll

Discover entities from the web using natural language:

from parallel_web_tools import run_findall

# Discover entities (auto-enriches by default)
result = run_findall("AI startups in healthcare", match_limit=20)

# Post-run operations
from parallel_web_tools import enrich_findall, extend_findall, get_findall_schema

schema = get_findall_schema(result.run_id)
enriched = enrich_findall(result.run_id, ["funding amount", "number of employees"])
extended = extend_findall(result.run_id, additional_matches=10)


# Fast, low-latency entity search (synchronous)
from parallel_web_tools import entity_search_findall

entities = entity_search_findall(
    "AI startups in healthcare",
    entity_type="companies",  # "companies" or "people"
    match_limit=25,
)

Monitor

Track web changes programmatically:

from parallel_web_tools import create_monitor, list_monitors, get_monitor

# Create an event_stream monitor (the default)
monitor = create_monitor(query="Track Tesla SEC filings", frequency="1d")

# List all monitors (cursor paginated)
monitors = list_monitors()

# Get monitor details
details = get_monitor(monitor["monitor_id"])

Memory

Retrieve and manage Memory entries:

from parallel_web_tools import clear_memory, evict_memory, retrieve_memory

# Relevant prior work across Task, Monitor, and FindAll
memories = retrieve_memory("serverless inference vendors", limit=10)

# Retrieve from an isolated memory scope
scoped = retrieve_memory(
    "inference vendors",
    memory_scope_key="workspace_acme",
)

# Remove one run from Memory; the underlying Task run remains available
evict_memory(
    "task",
    "trun_abc123",
    memory_scope_key="workspace_acme",
)

# Clear the selected memory; underlying runs remain available
clear_memory(memory_scope_key="workspace_acme")

These functions use the generated client.beta.memory resource provided by parallel-web 1.2.0 or later.

YAML Configuration Format

source: input.csv
target: output.csv
source_type: csv  # csv, json, duckdb, or bigquery
processor: core-fast  # lite, base, core, pro, ultra (add -fast for speed)

source_columns:
  - name: company_name
    description: The name of the company

enriched_columns:
  - name: ceo
    description: The CEO of the company
    type: str  # str, int, float, bool
  - name: revenue
    description: Annual revenue in USD
    type: float

Environment Variables

Related Packages

• parallel-web - Official Parallel Python SDK (this package depends on it)

Development

git clone https://github.com/parallel-web/parallel-web-tools.git
cd parallel-web-tools
uv sync --all-extras
uv run pytest tests/ -v

License

MIT