topoloop
#1 most accurate places API →

The places API built for AI agents

Global place search built to give agents spatial intelligence — matches Google on real agent queries, priced to be called in a loop, licensed to keep what you find. Your agent can finally afford to think.

Start building — free See what's different
10,000 free searches/mo No card required No surprise bills
Live demo agent → topoloop · POST /places/search
// your agent, mid-loop:
200 OK · 236 ms · top of 5
© Mapbox © OSM
Yari-ga-take
槍ヶ岳 · Hida Mountains, JP
Real agent tool-calls, resolved cross-language · descriptive · half-remembered — all top-rank
Place search

The most accurate place search an AI agent can call

Agents don't typo — they approximate: "the old lantern street," "haifoss waterfall," a romanization of 漢字. String matchers miss. topoloop learned place search from real agent traffic, so it resolves what the model means.

Any language in, any language out
Geo-anchored — never the wrong hemisphere
Descriptions work, no exact name needed
same query · three enginesactual responses
agent query
topoloop
Google
LocationIQ
"jøkulkyrkja summit" →
✓Kyrkja Jotunheimen, NO — #1
✗Jøkulkyrkja Mtn Antarctica — 14,000 km away
✗0 results empty response
"Aoba" →
✓旭川らうめん青葉 the original shop — #1
✗Ramen Aoba (branch) food court, not the original
✗Aoba Ward Sendai — wrong island
"black pepper pork bun" →
✓福州世祖胡椒餅 Raohe Night Market — #1
✓Fuzhou Pepper Buns found it
✗Pepperoni's Pizza North Carolina — 12,500 km away
right place in top 5 · all 650
0.93
0.93
0.62
judge-graded on 650 real agent tool-calls, July 2026 · full benchmark publishing soon
Rich place data

Get the place data Google refuses to sell

Google has web search and a places database — its API will never join them for you. topoloop ships them joined: verified official website, hours, menus, and web mentions from reputable sources, attached to every place. Your agent cites; it doesn't guess.

Official sites verified against the site itself
Hours, phone — and menus for restaurants
Web mentions, quoted and linked
place_detailsone call
茶
阿妹茶樓
A-Mei Teahouse · Jiufen, New Taipei
✓ official site verified hours + phone 5 blog mentions
A historic tea house in Jiufen, formerly a blacksmith shop — verified against its own website, confidence 0.98.
“Visit the famous A-Mei Lantern Tea House… the most picturesque one that you see in all the photos…”
girleatworld.net
“…staircase that leads directly up to Amei Teahouse (Jiufen's most famous attraction).”
nickkembel.com
Your data

Data you actually control

Other places APIs rent you answers — no caching, 30-day ID deletes, display only on their maps. topoloop results are yours to keep: cache forever, store IDs, train on them, export the lot.

Built on OpenStreetMap (ODbL) — no lock-in
Render on any map
Self-host if you outgrow the API
What you can do with results
topoloop
Google
Cache responses indefinitely
✓
✗
Store place IDs in your DB
✓
30 days
Display on any map (MapLibre, Mapbox…)
✓
✗
Use in training & evals
✓
✗
Bulk export your resolved data
✓
✗
Self-host the full stack
✓
✗
Pricing

Priced for loops, not sessions

Google charges $32 per thousand searches — pricing built for humans typing in boxes. Agents run loops, so we price for loops: true pay-per-call, hard caps on every tier, volume rates published all the way up. No "contact sales" cliff between you and scale.

what an agent spends on places calls
One research loop the agent searches 200 places, pulls details on 100
topoloop$0.36
Google Places$8.10
10,000 daily active users ~20 searches + 10 detail calls per user per day, for a month
topoloop$10k/mo
Google Places$243k/mo
list prices: topoloop $1.20/1k, $1.00/1k past 5M · Google $32/1k search + $17/1k details — up to 24× cheaper
Live now
Base Camp
$0 /mo
  • 10,000 calls / month
  • 2 requests / second
  • Search + nearby + details
  • MCP server access
  • Community support
Start free
Coming soon
Trailhead
$29 /mo
  • 25,000 calls / month
  • 50 requests / second
  • Enrichment fields included
  • Batch endpoints
  • Email support
Coming soon
Coming soon
Expedition
$199 /mo
  • 200,000 calls / month
  • Entity-resolution pipelines
  • 99.9% uptime SLA
  • Priority support
Coming soon
Summit / Self-host
Custom
  • Run the full stack inside your perimeter
  • Regional or planet bundles
  • Index + model refreshes
  • Data licensing available
Talk to us

The free tier is live today; paid tiers launch soon at these prices. Past 200k calls: $1.20 / 1k, $1.00 / 1k above 5M — published all the way up, no sales call.

Why choose us

The quadrant nobody else occupies:
buildable and good

Google has the quality but impossible pricing and terms. The cheap OSM tier is affordable but can't handle agentic queries. topoloop is both — here's the short version.

Google-grade accuracy

Trained on real agent traces: on our judge-graded benchmark the messy queries agents actually send match Google's hit rate — 31 points ahead of the cheap OSM tier.

A price that works for agents

True pay-per-call from $0: the research loop that costs $8.10 on Google costs 36¢ here. Hard caps, so a loop can never surprise your card.

License you can build on

Cache, store, train, export, display on any map. No 30-day deletes, no display-only clauses, no rug-pull in the terms.

MCP-native

One line adds topoloop to Claude, Cursor, or any MCP client — plus REST and typed SDKs that behave identically.

The whole planet, any language

49M places, 100+ languages, queried in whichever one your model thinks in. Ingested straight from the raw OSM planet file.

Fast enough for the loop

Sub-300 ms typical round trip, 50 rps from the first paid tier. Built to be called mid-reasoning, not batched overnight.

FAQ

Questions, answered straight

The things developers ask before their first key.

How is topoloop different from Google Places?
Three ways. Accuracy on agent queries: topoloop learned place search from real agent tool-calls, so approximate, descriptive, and cross-language queries resolve where string matchers miss — on our judge-graded benchmark of 650 real tool-calls it matches Google (R@5 0.93 vs 0.93) where the cheap OSM tier scores 0.62. Price: $1.20/1k flat vs Google's $32/1k search + $17/1k details — the agent loop that costs $8.10 there costs 36¢ here. License: Google forbids caching, limits place-ID storage to 30 days, and requires display on Google Maps — topoloop lets you cache, store, train, and render anywhere.
Is the free tier really free?
Yes — 10,000 calls a month, no card required. Every tier has a hard cap, so there is no scenario where usage silently rolls into a bill. When you outgrow it, paid tiers start at $29/mo with published volume rates all the way up.
Can I cache results, store IDs, and train models on them?
All three. Results are built on OpenStreetMap data (ODbL) plus our own enrichment layer, and the terms are written for builders: cache indefinitely, keep OSM place IDs in your database forever, use responses in training sets and evals, bulk-export what you've resolved.
How do I integrate it with my agent?
Fastest path is MCP: claude mcp add topoloop and your agent has search_places, nearby_places, and place_details as tools. There's also a plain REST API (two endpoints, obvious JSON) and typed Python and TypeScript SDKs, with LangChain and Vercel AI SDK bindings.
Where does the data come from?
The base layer is OpenStreetMap, ingested from the raw planet file — ~49M places, each carrying its OSM ID. On top of that sits our enrichment layer: verified official websites, hours, services, and descriptions, independently sourced and keyed to every place.
What happens if I outgrow the API?
You can't be stranded. Volume rates are published to 5M+ calls (down to $1.00/1k) with no sales call required, and if you need the index inside your own perimeter, self-host bundles ship the full stack — regional or planet-scale, with index and model refreshes.

Your agent already knows what.
Give it where.

10,000 free searches a month. One line of MCP. The whole planet, any language. Bring the loop — we'll bring the map.

Get your API key
36°20′31″N 137°38′51″E · CONTOURS: HIDA MOUNTAINS · INTERVAL 200 m