Why Pluriwave
Built for the lists that don’t exist yet
Analysts, investors and compliance teams spend weeks building lists by hand: finding every company, permit or listing that fits, then researching each one across dozens of websites and documents. Pluriwave does that work from a single sentence, at a scale single-agent tools can’t reach.
The list doesn’t exist yet
Finding the rows means searching many sources and removing duplicates. Pluriwave discovers candidates across sources and merges them into one row per company, clinic or listing.
Every cell is real research
Several lookups, a judgment call and a source someone can check. Each cell gets its own agent with web search, page fetching and a sandbox for data work.
It has to stay current
Schedule daily, weekly or monthly refreshes. Each one re-checks the table and reports exactly what changed.
Someone has to trust it
Every answer carries its evidence, a confidence level and its cost, so a client, an auditor or your own team can check the work.
How it works
You stay in charge at every gate
A run waits for you twice: once to approve the plan, once to approve the pilot. Nothing large happens without your say-so.
- 01
Plan
Write one sentence. The planner proposes what a row is, where to find rows, one column per check and an estimated cost, and it asks when something is ambiguous. Nothing runs until you approve.
- Row definition
- Sources to search
- One column per check
- Cost estimate
- 02
Pilot
About 30 rows run live. Open any cell to see the answer and the evidence behind it, then confirm it or correct it. Your corrections become a gold set.
- ~30 rows, live
- Answer + evidence
- Confirm or correct
- 03
Refine
The supervisor rewrites the checks that missed and re-tests every new version against your gold set, so a fix can’t quietly break what already worked.
- Versioned checks
- Tested on your labels
- No silent regressions
- 04
Run & monitor
Approve the full run under a budget cap and it fans out across thousands of rows. Filters drop rows that don’t fit before money is spent on them, and scheduled refreshes keep the table current.
- Budget cap
- Thousands of rows
- Scheduled refreshes
Features
Every answer, accounted for
Each cell is a small research project with its own agent, its own evidence and its own cost line. The table is the record.
Evidence on every answer
Sources, quoted passages and saved copies of the pages each answer relied on. “Unknown” is an allowed answer when the evidence isn’t there.
Cost before you commit
The plan estimates spend up front, every cell records what it cost, and a budget cap stops a run at your limit.
Filters that save money
Rows that fail a filter skip the remaining columns, and the table shows exactly which rule removed them and why.
Discovery across sources
Google Maps and Google Search through Apify, plus your own uploaded lists, merged into one row per entity.
Real tools in every cell
Each agent can search the web, read pages and run Python or SQL in an isolated sandbox when the answer needs computing.
The right model for each column
Claude models by default, matched to how hard each question is, with optional open-weight models and a switch to keep work away from providers in mainland China.
Versions and second opinions
Every check is versioned with its history. Add a second model’s opinion where a wrong answer would be expensive.
Yours to take anywhere
Export any table to CSV or Parquet, with the answer, confidence and sources for every cell.
Use cases
Any entity. Any field.
If you can describe the rows and the questions in a sentence, Pluriwave can build the table. A few lists you could start with:
“Find founder-owned HVAC contractors in Texas with 20 to 100 employees. Check union status and SBA loan history.”
“For every provider in our directory, confirm they still practice at the listed address and accept new patients.”
“Scan municipal and county procurement portals in Ohio, Indiana and Illinois for HVAC maintenance bids.”
“List rural health clinics in the Great Plains and match each one to open HRSA grants it qualifies for.”
“Find marketplace listings of our products by unauthorized sellers, with price, seller and listing URL.”
“Check each vendor against the state business registry: active status, formation date and registered agent.”
Security & trust
Built to be audited
Research you can hand to a client or a regulator has to show its work, and it has to keep each customer’s data apart. Both are part of the foundation, not features added later.
- Isolated by the database
- Each workspace’s data is separated by row-level security in Postgres itself, not only by application code.
- Sandboxed code, sealed secrets
- Agents run code in isolated sandboxes. Your connector secrets never enter them and are encrypted at rest with AES-256-GCM.
- A record of everything
- Every cell records its sources, model, attempts and cost. Operator actions are written to an append-only audit trail.
- Control where models run
- Exclude model providers based in mainland China for your whole workspace, with one switch.
Pricing
Pay for the research, not the seat
Credits cover model calls, searches, scraping and sandbox time. A credit is one cent, credits never expire, every plan comes with an estimate before anything runs, and every cell shows what it cost.
Starter
$10
1,000 credits
Try a pilot on your own list
- Every feature included
- Pilot before every full run
- Budget caps per run
Growth
$90
10,000 credits
Save 10%. Room for full runs
- Every feature included
- Pilot before every full run
- Budget caps per run
Scale
$800
100,000 credits
Save 20%. For tables that refresh
- Every feature included
- Pilot before every full run
- Budget caps per run
Running continuously?
Approved teams can switch to monthly invoicing for usage, with a credit limit instead of prepaid packs.
FAQ
Questions, answered
How is this different from asking a chatbot?
A chatbot answers one question in one conversation. Pluriwave builds a table: it finds the rows, runs an agent on each cell in parallel, keeps the evidence for every answer, and can re-run the whole table on a schedule and tell you what changed.
How accurate are the answers?
Every answer links to its evidence, carries a confidence level, and may honestly be “unknown”. You review a pilot before the full run, and your corrections become a gold set that every revised check is tested against before it’s used.
What will a run cost?
The plan includes an estimate before anything runs, the pilot shows real costs on about 30 rows, and you set a budget cap for the full run. Every cell shows exactly what it cost, and filters stop spending on rows that don’t fit.
Where do the rows come from?
Discovery searches sources such as Google Maps and Google Search through Apify, and you can upload your own lists. Candidates from every source are matched and merged into one row per entity.
Which AI models do you use?
Claude models by default, chosen per column for how hard the question is. You can enable additional open-weight models, and exclude providers based in mainland China for your workspace.
Can the table stay up to date?
Yes. Schedule daily, weekly or monthly refreshes. Each refresh re-checks the table and reports what changed since the last one.
Can I export the results?
Yes. Download any table as CSV or Parquet, with answers, confidence and sources for every cell.
How do I pay?
Buy credit packs as you go: a credit is one cent, and larger packs cost less per credit. Teams that run continuously can apply for monthly invoicing instead.