Exa Agent API

The same job as one request

An Exa Agent run takes a query and an effort. A SpecStream request takes the job in plain words and answers with the price first.

# Exa Agent: a run at an effort, priced by the effort or metered to a cap
curl -X POST https://api.exa.ai/agent/runs \
  -H "x-api-key: $EXA_API_KEY" -H "Content-Type: application/json" \
  -d '{"query": "Engineering leaders at AI infrastructure companies that raised a Series A or B in the last 6 months", "effort": "auto"}'

# SpecStream Deep Search: the same job, read back with its price before anything runs
curl -i -X POST https://api.specstream.ai/v1/agent/lists \
  -H "Content-Type: application/json" \
  -d '{"request": "Engineering leaders at AI infrastructure companies that raised a Series A or B in the last 6 months, with their title and company", "count": 25}'

The first reply is 402 Payment Required. Its body shows how SpecStream read the request, the columns and the most it can cost, and nothing is spent. Pay that maximum and send the same request again: you get a status_url, and the rows are there when the run ends.

A price per row, approved before the run

Exa prices a fixed effort per run, whatever the run finds, and meters auto and ultra in compute units and searches up to the cap. On a fixed effort, budget.maxCostDollars limits only contact enrichment, and Exa's docs say the cap does not limit research or reasoning steps. SpecStream bills the rows it delivers. At 5¢ a row and 5¢ for each work email, 200 companies with an email cost $20, a result that fails a condition is not billed, and the most the run can cost is approved before it starts.

Source URL and read time on every value

Exa returns grounding for text or structured fields when it emits it. On SpecStream each row carries the source URL and the time it was read, and each condition of the request shows the page that states it, or that no page states it yet. SpecStream reads each page and joins what it finds into one record per company or person.

The rows stay up to date in your tools

Save a request and it reruns hourly to yearly, inside the spending limits you set. Each row comes back labeled changed, unchanged or no longer matching. A changed or new row costs half its rate, an unchanged row costs 1¢, and new matches land in HubSpot, Salesforce, Pipedrive, Attio, Sheets, Slack or Teams, or in a signed webhook.

From your agent over MCP

Connect https://api.specstream.ai/mcp over Streamable HTTP, with OAuth or a SpecStream key. compile reads the request into editable parts, quote_capacity_work prices it, authorize_capacity_work runs it inside the maximum your person approved, and list_find_rows reads the rows with their sources. The MCP guide covers each client.

The most accurate web search API on the same key

On SimpleQA Verified the SpecStream Search API scores 97.9 F-score on all 1,000 official questions, 978 of them correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks. The Extract API scores 0.908 F1 on WCXB. Every score and its method is on the benchmarks page.

Parallel Task API and FindAll

Task runs are priced per 1,000 by processor: $5 on lite and $300 on ultra, which is 30¢ a run. FindAll costs $0.25 a run plus $0.03 a match on base. On SpecStream the price is set per delivered row and quoted for the whole request before it starts.

Tavily Research

A research endpoint billed in credits at $0.008 each pay as you go: 4 to 110 credits a request on mini and 15 to 250 on pro, so 3¢ to $2 a request. SpecStream delivers rows, each value with its source URL and read time.

Perplexity Agent API

Billed by model tokens plus tool calls, with web search at $2.50 per 1,000 calls; Perplexity's pricing page says effort costs vary with the model, token usage and tool invocations. SpecStream names the most a run can cost before it runs.

Exa Deep Search

The deep search type costs $12 per 1,000 requests and returns synthesized output from one search call. The SpecStream Search API returns the answer with the page it was read from at $5 per 1,000, and Deep Search returns every match as rows.