Smart Search
A direct, cited answer to any question — synthesized in one call.
Early access · Now in public betaThe search layer
The search layer
for your AI agents.
Live web search, smart answers, and deep research in a single, predictable API. Designed for builders who'd rather ship features than scrapers.
1,000 free credits / month · No card · Cancel anytime
Smart Answers
One question. One cited answer.
Skip the ten-blue-links workflow. Send a question, get a clean synthesized answer back with every claim linked to its source — streamed token by token, ready to display.
- Real-time synthesis from up to 20 sources
- Inline numbered citations, every time
- Streams over Server-Sent Events
- 2 credits per call. That's it.
Deep Research
From topic to structured report in one call.
Our research engine plans sub-questions, runs parallel searches, ranks every source, extracts clean content, and returns a long-form report with full citations. You watch it work in real time.
- 4–5 planned sub-questions per topic
- AI-ranked sources across the pool
- Markdown report with sections + bullets
- Cached for 24h — repeat queries are free
One HTTP call. Any stack.
Plain JSON over HTTPS — works from any language. Authenticate with a Bearer token from your dashboard.
curl 'https://api.pellumin.com/api/summary?q=tls+handshake' \ -H "Authorization: Bearer plm_live_…"
One API. Every web primitive your agent needs.
Pick the depth you need — from raw ranked links to a full cited research report. Read the full reference →
Deep Search
Multi-step research reports with planning, ranking, and full source trails.
Web Search
Raw ranked web results — optional AI relevance scoring per call.
Extract
Any URL → clean Markdown. Boilerplate, ads, and chrome stripped.
Crawl
BFS-crawl an entire domain at scale, respecting robots.txt.
Security firewall
Every fetched page is sanitized: prompt injection stripped, PII redacted.
Teams shipping AI products.
From a chat widget to a fully autonomous agent.
Autonomous AI agents
Give your agent a real research tool — plan, search, extract, synthesize.
AI chat & copilots
Ground answers in the live web so your assistant always has fresh context.
Content & SEO ops
Brief writers with structured, cited intelligence from the open web.
Customer support
Surface the right answer from public docs and tickets in milliseconds.
Market intelligence
Monitor competitors, news, and pricing continuously.
Internal tools
Power dashboards and briefings without scraping yourself.
Stop scraping. Start shipping.
Roll your own
With Pellumin
Time to first call
Days of scraping + parsing
Five minutes, two lines of code
Result quality
Raw HTML you have to clean
Ranked, cited, ready for an LLM
Maintenance
Selectors break weekly
We keep it running
Cost predictability
Mystery infra bill
Flat credit price per call
AI safety
Inject anything into your model
Built-in prompt-injection firewall
Three things we won't compromise on.
We're an early-stage team building this in the open. These are the principles we ship against.
Predictable by design
Every operation costs a known number of credits. Forecast your bill before you ship.
Safety in the box
We strip prompt injection and redact PII from every web response — before it reaches your model.
Built to drop in
First-class SDKs, framework integrations, and an MCP server. Use whatever stack you love.
From signup to first call in under five minutes.
Sign up
Create your workspace in 30 seconds. We'll seed it with 1,000 free credits every month.
Generate a key
One click in the dashboard. Treat it like a password — anyone with it can spend.
Ship
Two lines of code. The Playground sandboxes every endpoint before you go live.
Quick answers.
Do I need a credit card to start?
No. Sign up gives you 1,000 credits every month — free, no card on file.
What counts as one credit?
A Smart Search costs 2 credits — one cited synthesized answer. A Deep Search costs 10 credits — a full multi-step research report. Free-plan credits reset every month.
Which frameworks can I use?
Python, TypeScript, LangChain, Vercel AI SDK, and Model Context Protocol. Or just hit the REST API directly.
How fresh are the results?
Real time. Every query hits live sources at request time and returns ranked, deduped results.
Is the content safe to feed my model?
Yes — every fetched page passes through a firewall that strips prompt-injection attempts, redacts PII, and flags unsafe URLs.
Are you a real company?
We're an early-stage startup. The product is real and live; the team is small. Sign up free and put it to work — we'd love your feedback.
Ship a smarter agent this week.
Get your workspace and 1,000 credits in 30 seconds.
Create my workspace →