> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getcargo.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Parallel

> Connect Parallel AI for web search, content extraction, and deep research tasks in your workflows.

## How to set up Parallel

### Authentication

Parallel runs on Cargo credits out of the box, so you can use it without a Parallel account. To bill Parallel directly instead, connect your own API key:

1. Sign up at [parallel.ai](https://parallel.ai)
2. Create an API key in your Parallel dashboard
3. Paste the API key in Cargo when connecting

<Note>
  **Code slugs** — integration slug: `parallel` · actions: `search`, `extract`, `createTask`. In a [workflow](/workflows/overview), declare the connector in `uses` and call `uses.<key>.search(...)`.
</Note>

## Parallel actions

### Search

Search the web with Parallel AI and return ranked results with relevant excerpts.

**Required fields:**

* **Search queries**: Concise keyword queries, 3-6 words each; provide 2-3 for best results

**Optional fields:**

* **Objective**: Natural-language description of the question or goal driving the search
* **Search mode**: `basic` for lowest latency, `advanced` for higher quality (default)
* **Number of results**: Maximum results to return (1-100)
* **Max characters total**: Upper bound on total characters across excerpts
* **Include domains**: Only return results from these domains (accepts `.gov`-style extensions)
* **Exclude domains**: Exclude results from these domains
* **After date**: Only include results published on or after this date (YYYY-MM-DD)

**Use case:** Find recent news, funding announcements, or public signals about an account before outreach.

### Extract

Extract relevant content from specific web URLs.

**Required fields:**

* **URLs**: URLs to extract content from (up to 20)

**Optional fields:**

* **Objective**: Natural-language goal used to focus excerpts on the most relevant content
* **Search queries**: Keyword queries used together with the objective to focus excerpts
* **Max characters total**: Upper bound on total characters across extracted excerpts
* **Include full content**: Also return the full page content as markdown

**Use case:** Pull pricing, careers, or product details from known pages of a prospect's website for enrichment.

### Create Task

Execute a full web research task — complex queries that require deep research, analysis, and structured output. The task runs asynchronously on Parallel's side and the workflow resumes when it completes (up to 60 minutes).

**Required fields:**

* **Input**: A natural-language question or a description of what to research
* **Processor**: `lite`, `base`, `core` (balanced, default), `core2x`, `pro`, `ultra`, `ultra2x`, `ultra4x`, or `ultra8x` — from fastest and cheapest to deepest and most expensive

**Optional fields:**

* **Output schema**: Description of the desired output — natural language or a JSON schema as a string
* **Include domains**: Only use results from these domains
* **Exclude domains**: Exclude results from these domains
* **After date**: Only include results published on or after this date (YYYY-MM-DD)
* **Location**: ISO 3166-1 alpha-2 country code for geo-targeted results (e.g., `us`, `gb`)

The result includes the task output, output metadata (citations and confidence), and the sources used.

**Use case:** Research an entire account — business model, ICP fit, recent initiatives — and return it as a structured object for scoring.

## Best practices

* Use Search for quick lookups and Create Task for research that needs synthesis across many sources
* Give Create Task an output schema to get structured, mappable results instead of free text
* Scope searches with include/exclude domains and an after date to cut noise and stale results
* Pick the cheapest processor that does the job: `lite` or `base` for simple lookups, the `ultra` tiers only for genuinely complex questions — each step up the ladder multiplies the credit cost
* Cap **Number of results** on Search: the first 10 come with the request, and every result past that adds to the bill
* Account for Create Task's asynchronous nature (up to 60 minutes) when designing time-sensitive workflows

## Credits

Parallel is billable with Cargo credits, so you can use it without your own Parallel account. Costs mirror Parallel's published pricing at \$0.04 per Cargo credit.

* **Search**: 0.125 credits per search, which covers the first 10 results, plus 0.025 credits for each additional result returned
* **Extract**: 0.025 credits per URL submitted, whether or not the page could be read
* **Create Task**: a flat cost per run that depends on the processor

| Processor | Credits per task |
| --------- | ---------------- |
| `lite`    | 0.125            |
| `base`    | 0.25             |
| `core`    | 0.625            |
| `core2x`  | 1.25             |
| `pro`     | 2.5              |
| `ultra`   | 7.5              |
| `ultra2x` | 15               |
| `ultra4x` | 30               |
| `ultra8x` | 60               |

Cargo tracks credit usage and supports caching to optimize costs.

## Rate limits

Parallel rate-limits each endpoint separately, and only counts the requests that create work — retrieving a task result is free. Cargo spreads calls to stay under the limits and retries on rejection.

* **Search**: 600 requests per minute
* **Extract**: 600 requests per minute
* **Create Task**: 2,000 task runs per minute

On Cargo credits these limits are shared across all workspaces on Cargo's key; on your own API key they apply to your connector alone.