Web infrastructure for AI agents · v1.0

Make your agents 1000× more powerful.

One API that gives agents and apps the live web: search, deep research, extraction and monitoring. Every answer comes back structured, and cited to the sentence it came from.

2,481,000,000pages read for agents today
48 msp50 search latency
74%deep-research accuracy
1 callsearch → read → cite
How it works

One call. Four moves your agent can't make alone.

Every request runs the same pipeline across thousands of pages in parallel, and returns only what your agent needs to act.

Search01 / 04

Find

Turns an objective into dozens of queries across an index refreshed every few minutes, then ranks by usefulness to an LLM, not clicks.

› objective: "climate-tech Series B, India, 2026"
expanding to 14 queries…

inc42.com/funding/varaha-series-b0.97
economictimes.com/tech/ather-grid0.95
yourstory.com/2026/04/kirana-solar0.91
livemint.com/companies/log9-cells0.88
+ 408 more candidates
Read02 / 04

Read

Renders JavaScript, clears cookie walls and paywalled previews you license, and converts each page into clean, chunked text.

1,284 pages · 3.1M tokens → 41k useful

# Varaha raises $42M Series B
Varaha, the Delhi-based carbon removal
company, raised $42 million in a round
led by Omnivore, with participation…

— stripped: nav, ads, 14 related links
Reason03 / 04

Reconcile

Cross-checks every claim across sources, resolves conflicting figures and flags what it couldn't verify instead of guessing.

round_size("Ather Grid")
├ inc42 ········ $88M
├ et.tech ······ $88M
├ press release · $88M
└ blog post ···· $80M (stale, 2025)

resolved → $88M · confidence 0.97
Cite04 / 04

Return

Gives your agent typed rows that match your schema, with a citation on every field down to the sentence.

{
  "company": "Varaha",[1]
  "round": "Series B",[1]
  "amount_usd": 42000000,[1][2]
  "lead": "Omnivore",[2]
  "confidence": 0.94
}
Task API · live

Hand it a week of analyst work.

Pick a processor and press run. ActionLayer plans the sub-queries, reads in parallel and streams rows back as they're verified.

task_01JD8Q2
Plan
    CompanyFounderRoundLeadSourcesConf.
    pages read 0citations 0elapsed 0.0scost $0.000
    Benchmarks

    Measured on the questions agents fail.

    Accuracy on 1,266 multi-hop web research questions, graded against hand-verified answers. Same questions, same time budget.

    3.1×more correct answers than a frontier model browsing alone
    48msp50 Search latency, query to ranked excerpts
    $0.004per Base task, billed per task, never per token
    Processors & pricing

    Choose depth per call.

    Every endpoint takes a processor. Spend milliseconds on a lookup and minutes on a due-diligence memo, through the same API.

    ProcessorBest forDepthLatencyPer 1k calls
    LiteFast lookups inside a chat turn~50 ms$1
    BaseExtraction to your schema~1.2 s$4
    ProMulti-hop research with citations~45 s$100
    UltraLong tasks, forms, monitoring2–30 min$300
    Docs

    Four lines from prompt to proof.

    • TypedPass a schema and get validated JSON with a citation on every field.
    • MCPOne server for Claude, Cursor or your own agent loop.
    • StreamsWatch tasks think, or get a webhook when the web changes.
    
        
    Start building

    Give your agents the whole web.

    10,000 free Search calls and 100 Pro tasks every month. No card required.

    Frequently asked questions

    What is ActionLayer?

    ActionLayer is one API that gives AI agents and apps the live web: search, deep research, extraction and monitoring, returned as structured data.

    How is the output structured?

    You choose the depth per call. Answers come back as structured data with citations back to the sentence they came from.

    Which tasks does ActionLayer cover?

    Four moves: find results, read pages, reconcile sources and return an answer, with monitoring for changes over time.

    Is ActionLayer available now?

    ActionLayer is in pre-launch. Join the waitlist for launch access and updates.