Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.parallel.ai/llms.txt

Use this file to discover all available pages before exploring further.

For AI agents: a documentation index is available at https://docs.parallel.ai/llms.txt. The full text of all docs is at https://docs.parallel.ai/llms-full.txt. You may also fetch any page as Markdown by appending .md to its URL or sending Accept: text/markdown.
TermDefinition
AgentAn AI system that uses tools and external data sources to autonomously plan, decide, and execute multi-step tasks. Parallel’s APIs —Search, Extract, Task, FindAll, Monitor, and Chat — are designed to give agents reliable, citation-grounded access to the web.
Agent SkillsLightweight, declarative integrations that add Parallel’s web search, extraction, deep research, and data enrichment capabilities to AI coding agents like Cursor, Cline, Claude Code, and 30+ other tools. See Agent Skills.
Agentic PaymentsA payment model that lets AI agents autonomously pay for Parallel APIs using the Machine Payments Protocol (MPP) via Stripe or Tempo stablecoins, or via the x402 protocol over HTTP 402 Payment Required. See Agentic Payments.
API keyA unique identifier used to authenticate requests to Parallel APIs. Generate your API key at platform.parallel.ai.
Asynchronous APIAn API that returns a run ID immediately and processes the request in the background. Task, FindAll, and Monitor APIs are asynchronous. Poll for status or use webhooks to receive results.
CandidateA potential entity discovered during a FindAll run. Candidates are generated from web data and evaluated against match conditions.
Chat APIA synchronous API that provides OpenAI-compatible streaming chat completions with web-grounded responses. Supports both speed-optimized and research-grade models.
CitationA reference to a web source that contributed to an output field. Includes the URL and relevant excerpts from the source.
CLIThe parallel-cli command-line tool for invoking Parallel’s Search, Extract, Task, FindAll, and Monitor APIs from a terminal or agent. Designed for use in scripts, headless environments, and standalone agents. See Parallel CLI.
Confidence levelA reliability rating for each output field: high (strong evidence from multiple authoritative sources), medium (adequate evidence with some inconsistencies), or low (limited or conflicting evidence).
CrawlerA bot that systematically fetches and indexes pages from the web. Parallel operates ShapBot, the crawler that powers its web index and Search API. See Crawler.
EnrichmentAdditional structured data extracted for matched candidates using the Task API. Enrichments run automatically on candidates that pass all match conditions.
EventA detected change or update that matches a monitor’s query. Events include the detected information, event date, and source URLs.
Event groupA collection of related events detected during a single monitor execution.
ExcerptA focused portion of page content that’s relevant to your objective. Optimized for LLM consumption.
Extract APIA synchronous API that converts any public URL into clean, LLM-optimized markdown. Handles JavaScript-heavy pages and PDFs.
Fast processorA processor variant optimized for speed over data freshness. Append -fast to any processor name (e.g., core-fast) for 2-5x faster response times.
FieldBasisThe specific object containing citations, reasoning, and confidence for an individual output field within the research basis.
FindAll APIAn asynchronous API for web-scale entity discovery. Turns natural language queries into structured, enriched databases by generating candidates, validating them against criteria, and optionally enriching matches.
FindAll runA single execution of a FindAll query. Includes candidate generation, evaluation, and optional enrichment.
FrequencyThe frequency at which a monitor executes. For example: 1h (hourly), 2d (every other day), 4w (every four weeks).
GeneratorThe engine that determines the quality and thoroughness of FindAll run results. Options include preview, base, core, and pro.
IndexerThe system that processes pages fetched by the crawler and stores them in a structured, queryable index. Parallel’s indexer powers our API coverage across 30+ countries and specialized verticals like coding, company, and finance search.
Ingest APIAn API that generates Task specifications (input/output schemas and processor recommendations) from a natural-language description of an objective. See Ingest API.
interaction_idAn identifier returned on every Task API and Chat API response. Pass it as previous_interaction_id on a subsequent request to chain context across calls and build multi-turn research workflows. See Interactions.
Match conditionA criterion that candidates must satisfy to be included in FindAll results. Defined in natural language as part of the query.
Match statusThe state of a candidate after evaluation: matched (satisfies all conditions), unmatched (fails one or more conditions), or generated (not yet evaluated).
MCP (Model Context Protocol)A protocol for connecting AI models to external tools and data sources. Parallel provides MCP servers for Search and Task APIs.
MonitorA scheduled query that continuously tracks the web for changes relevant to a specific topic.
Monitor APIAn asynchronous API for continuous web tracking. Creates scheduled queries that detect relevant changes and deliver updates via webhooks.
ObjectiveA natural language description of what the Search API (or Extract API) should focus on when retrieving results — for example, "Find recent FDA approvals for GLP-1 drugs". The Search API uses the objective to rank and compress excerpts so they directly address the question, rather than just matching keywords.
OrchestrationThe coordination of multiple agent steps, tool calls, and API requests into a single end-to-end workflow. Parallel handles orchestration internally—for example, FindAll automatically runs Task API enrichments on matched candidates, and the Task API harness orchestrates search, extraction, and reasoning across multiple sources.
Previous interaction IDThe interaction_id from a prior Task or Chat API call, passed on a new request to carry forward context. See interaction_id.
ProcessorThe engine that executes Task Runs. Processors vary in performance characteristics, latency, and reasoning depth. Options include lite, base, core, core2x, pro, ultra, ultra2x, ultra4x, and ultra8x. Each is available in standard and fast variants.
RankerThe component of a search engine that scores and orders candidate pages returned by the indexer to produce final results. Parallel’s ranker is tuned for agent-friendly retrieval, prioritizing pages that answer the objective with citation-ready evidence.
Rate limitThe maximum number of API requests allowed within a time period. See Rate limits for default quotas.
Research BasisThe structured explanation detailing the reasoning and evidence behind each Task Run result. Includes citations, reasoning, and confidence levels.
run_idThe unique identifier returned by the Task API when a Task Run is created (prefixed trun_). Use it to poll for status, retrieve results, stream events, or cancel the run.
Search APIA synchronous API that executes natural language web searches and returns LLM-optimized excerpts. Replaces multiple keyword searches with a single call for broad or complex queries.
Search modeA preset that configures Search API behavior. basic optimizes for low latency in foreground agents; advanced (default) uses a more thorough retrieval and compression pipeline for background agents. See Search Modes.
Search queryA specific search term or phrase used to find relevant pages. Multiple search queries can be combined in a single Search API request.
SDKSoftware Development Kit. Parallel provides official SDKs for Python and TypeScript.
ShapBotThe user-agent name of Parallel’s web crawler. See Crawler for the full user-agent string and IP ranges.
Snapshot monitorA monitor that re-runs a Task Run on a schedule and fires a webhook when the output materially changes. Structured JSON outputs are diffed field-by-field; text outputs are diffed as a whole. See Create a Snapshot Monitor.
Source policyConfiguration that controls which web sources can be accessed during research. Can include or exclude specific domains.
Synchronous APIAn API that returns results immediately in the response. Search, Extract, and Chat APIs are synchronous.
Task APIAn asynchronous API that combines AI inference with web search and live crawling to turn complex research tasks into repeatable workflows. Returns structured outputs with citations and confidence levels.
Task GroupA collection of Task Runs that can be executed and tracked together. Useful for batch processing multiple inputs.
Task RunA single execution of a task specification. Each Task Run processes one input and produces one output with its associated research basis.
Task specThe definition of what a Task Run should accomplish. Includes input schema, output schema, and optional instructions.
WebhookAn HTTP callback that delivers notifications when specific events occur (e.g., task completion, monitor event detection).
Zero Data Retention (ZDR)A deployment mode in which Parallel does not retain request or response data after a run completes. Interactions and other stateful features that depend on retained context are unavailable to ZDR customers.