Navi is joining Techstars Boulder 2026 and closing a $500K pre-seed. Read the post

Other maps know where things are. Navi knows what they're like.

Scored streets, real reads on every area, and maps drawn for people on foot, so your agent can send someone somewhere with confidence.

Request early access See what your agent gets back

Try the sandbox

// one call
const take = await navi.take({
ask, location, time, constraints
})

marlowe

hollis

navi's take

Go to Hollis, it's the quiet one.

leave by 2:52 Walk it, 4 min: faster than a car at pickup time.

Illustrative example. Places, quotes and times are fictional.

Why Navi

Other maps measure the distance. Navi knows the vibe.

Search, routes, and geocoding are in the box. What's different is what's underneath.

01 · every street, scored

Not just how far. How it feels.

85M+ street segments scored across 7 dimensions: day, night, safety, accessibility, and more. Your agent knows the difference between a lit side street and a dark shortcut, not just that one is 40 seconds faster.

By day, the shortcut on 2nd is fine.

2nd St · Pine to Oak1 segment · 7 scores

PineOak

lighting 82 traffic 74 safety 80 accessibility 70 activity 68 noise 72 sidewalk 76

02 · what this area is really like

The read a local would give you.

Busy with the office crowd at lunch, quiet by 7, lively again after 9. What a block, a neighborhood, or a place is for, who it's for, and when, with a source behind every claim.

North End.weekdays

morningcommuters, calm

lunchbusy, office crowd

eveningquiet by 7

latelively after 9

"Empties out after 7, then the bars on Pine fill up around 9."from reddit

03 · directions that choose the street

Maps built for people, not cars.

Walk, transit, drive, or bike, weighed on how the streets feel at that hour, not just the shortest line. Every route says why it wins and what to switch to if it rains or runs late.

Getting there · walk it, 4 min

Faster than a car at pickup, school traffic backs up 5th.

Raining: rideshare, 9 min.

04 · maps drawn for people

Maps that show how streets feel.

Any map can be styled. Navi's tiles carry what's underneath: every street's score and every area's character, drawn in, with the route Navi chose. Embed it or send the link.

somewhere quiet near the dentist for my call?

Here's the plan.

marlowe

hollis

leave by 2:52

EmbedCopy link

Coming next: Navi watches the plan and tells your agent when something changes, and plans across several people at once.

Illustrative example. Places, quotes and times are fictional.

What your agent gets back

Same ask. Two different maps.

the ask

User has a 2pm that may run long and a 3pm dentist appointment. find somewhere near the dentist to take a call. quiet, walkable, not a chain.

a typical mapping api

brewline coffee co. 4.4 ★ 0.1 mi · 1 min drive café

the daily cup 4.2 ★ 0.2 mi · 1 min drive café

marlowe 4.6 ★ 0.2 mi · 2 min drive café

bean & leaf 4.1 ★ 0.3 mi · 2 min drive coffee shop

hollis 4.5 ★ 0.3 mi · 2 min drive café

north end bakery 4.3 ★ 0.4 mi · 3 min drive bakery

corner grind 3.9 ★ 0.4 mi · 3 min drive coffee shop

the reading room 4.7 ★ 0.6 mi · 4 min drive library café

fifth & pine 4.0 ★ 0.7 mi · 4 min drive restaurant

A typical mapping API sees distance and drive time. Your agent still has to guess what any of these are like.

navi

{
  "plan_check": {
    "works": true,
    "slack_min": 8,
    "leave_by": "14:52",
    "note": "4 min walk to the dentist, open until 6"
  },
  "recommendation": "hollis",
  "area_read": {
    "area": "north end",
    "now": "quiet after the lunch crowd, lively again after 9",
    "for_whom": "people working solo in the afternoon",
    "sources": [{ "type": "reddit", "quote": "..." }]
  },
  "getting_there": {
    "mode": "walk",
    "minutes": 4,
    "why": "faster than a car at 2:50, school pickup backs up 5th",
    "instead": [
      { "mode": "rideshare", "minutes": 9, "if": "it's raining" },
      { "mode": "transit", "minutes": 11, "if": "you're coming from the 2pm across town" }
    ]
  },
  "picks": [{
    "id": "plc_8f2k",
    "name": "hollis",
    "for_whom": "people working solo, quiet after 1pm",
    "why": "spaced-out tables, quiet enough for a call",
    "walk": { "minutes": 4, "read": "flat, lit, low-traffic side street" },
    "sources": [{ "type": "reddit", "quote": "..." }]
  }],
  "tradeoffs": "the corner spot is closer but loud with a line at 2pm",
  "view": {
    "plan_id": "pln_7c1q",
    "url": "https://navi.app/v/pln_7c1q",
    "embed": "<navi-view plan="pln_7c1q">"
  },
  "constraints_held": ["near dentist", "quiet", "not a chain", "back by 3:00"]
}

a typical map

5 min · shortest

navi

6 min · lit, quiet, low traffic

calmbusy

One extra minute. A much better walk.

the street, scored

walk.read

the area, read

area_read

the route, chosen

getting_there

the map, drawn

view

Illustrative example. Places, quotes and times are fictional.

Compare

Other maps see the city like a car does. Navi sees it on foot.

The major mapping and places providers now offer tools for agents. Here's what your agent gets back from each.

places api, mcp server, recommendations

maps grounding, grounding lite mcp

mcp server

places api, mcp

a model with web search

built around people on foot search and driving vehicles and navigation venues whatever it finds

what a call returns scored streets, a read on the area, a chosen route, a drawn map, and the pick places, routes, weather maps, place search, routing venues, details, tips, photos text and links

what it knows about a place or area what it’s like, for whom and when, from what people say ratings, reviews, summaries name, category, address; movement data sold separately for analytics tips, attributes, hours whatever it finds

the walk scored on how the streets feel: day, night, safety, accessibility distance and travel time walking directions, distance and time not offered an estimate

sources a source on each claim links back to maps n/a user tips links, sometimes

agent-ready preview, mcp or http yes, managed mcp yes, mcp server (account and token required) yes, mcp server built in

built around

navi

people on foot

googlesearch and driving mapboxvehicles and navigation foursquarevenues web searchwhatever it finds

what a call returns

navi

scored streets, a read on the area, a chosen route, a drawn map, and the pick

googleplaces, routes, weather mapboxmaps, place search, routing foursquarevenues, details, tips, photos web searchtext and links

what it knows about a place or area

navi

what it’s like, for whom and when, from what people say

googleratings, reviews, summaries mapboxname, category, address; movement data sold separately for analytics foursquaretips, attributes, hours web searchwhatever it finds

the walk

navi

scored on how the streets feel: day, night, safety, accessibility

googledistance and travel time mapboxwalking directions, distance and time foursquarenot offered web searchan estimate

sources

navi

a source on each claim

googlelinks back to maps mapboxn/a foursquareuser tips web searchlinks, sometimes

agent-ready

navi

preview, mcp or http

googleyes, managed mcp mapboxyes, mcp server (account and token required) foursquareyes, mcp server web searchbuilt in

Based on each provider's public documentation as of september 2026. product names are trademarks of their owners.

The calls

The calls your agent can make.

The mapping calls you'd expect, plus the ones that know the streets. A preview of the surface, shaped as tools over MCP or HTTP.

navi.area_read(area | location, time?)                 // what this area is like, for whom, and when, sourced
navi.route(from, to, mode?, feel?)                      // walk | transit | drive | bike, weighed on how the streets feel
navi.take(ask, location, time, party?, constraints?)    // the right pick, the trade-off, how to get there
navi.view(plan_id)                                      // the map on navi tiles, ready to embed or link
navi.place_read(place_id, time?)                        // for what, for whom, when, sourced
navi.search(query, near, open_at?)                      // places, ranked for this ask
navi.geocode(address)                                   // address to a place, and back
navi.share(plan_id)                                     // a link a person or another agent can open
navi.check(plan)                                        // will it work, and the fix if not
navi.plan(stops[], window)                              // stops sequenced with leave-by times
navi.watch(plan_id, webhook)                            // next: an event when something changes
GET  /v0/areas/read?near=&time=                         // what this area is like, for whom, and when, sourced
POST /v0/route        { from, to, mode?, feel? }        // feel: balanced | night | safety | accessibility
POST /v0/take         { ask, location, time, party?, constraints? }
GET  /v0/view/:plan_id                                  // the map on navi tiles, embed or link
GET  /v0/places/:id/read?time=                          // for what, for whom, when, sourced
GET  /v0/search?q=&near=&open_at=                       // places, ranked for this ask
GET  /v0/geocode?address=                               // address to a place, and back
POST /v0/share        { plan_id }                       // a link a person or another agent can open
POST /v0/check        { plan }                          // will it work, and the fix if not
POST /v0/plan         { stops, window }                 // stops sequenced with leave-by times
POST /v0/watch        { plan_id, webhook }              // next: an event when something changes
Take {
  recommendation: PlaceId
  area_read: { area, now, for_whom, sources[] }
  getting_there: { mode, minutes, why, instead[] }  // walk | rideshare | transit | drive | bike
  picks: Pick[]                 // one to three, never padded
  tradeoffs: string
  view: { plan_id, url, embed }
}
Pick { id, name, for_whom, why, walk: { minutes, read }, sources: Source[] }

core

navi.area_read

What this area is like, for whom, and when, sourced

core

navi.route

Walk, transit, drive or bike, weighed on how the streets feel at that hour

core

navi.take

The right place for this person, right now, and how to get there

core

navi.view

The map on Navi’s own tiles, ready to show your user

core

navi.place_read

What a place is for, for whom, and when, with sources

basics

navi.search

Places near an ask, ranked for it, not just by distance

basics

navi.geocode

An address to a place, and back

basics

navi.share

One plan several people (or agents) can open

plans

navi.check

Does this plan actually work, and what’s the fix

plans

navi.plan

A plan that holds: order, timing, buffers

navi.watch

Tells your agent when the plan needs to change

Where it fits

For the agents people already use.

personal assistants

Fit a call into the gap between meetings, and make it back on time

scheduling agents

Pick where to meet so nobody’s late

travel and concierge agents

The right thing near me, right now

group-planning agents

One plan four people can say yes to

FAQ

The questions you'd ask.

Yes. Search, geocoding and routes are in the box. What’s different is underneath: every street scored on how it feels to walk, a read on what each area is like, and maps drawn to show both. If all you need is coordinates, any mapping API works.

It’s a preview. We’re opening early access to a small group first.

The scored streets come from what Navi already does in its own app. Watching a plan and telling your agent when something changes is next, and so is planning across several people at once.

Search, routes and the map view work wherever the underlying place data does. The scored streets and the local reads cover select cities today and grow city by city. Tell us where you need it.

Pricing isn’t set. Early access partners help shape it.

Yes. Every answer names the mode (walk, rideshare, transit, drive or bike), why it wins at that hour, and what to switch to if it rains or runs late. Navi recommends the mode; it doesn’t book the ride.

Yes. Every answer comes with a map on Navi’s own tiles: street scores, the area and the route Navi chose. Embed it or send a link. You don’t build a map stack.

No. Navi works alongside whatever you use for coordinates and basemaps. Call Navi when the walk, the area or the pick matters.

Any. It’s a tool call over MCP or HTTP.

Tell us what you need in the form. Terms are being set with early partners.

The design is that you pass preferences per call and keep your own memory layer.

Early access

Tell us what your agent needs.

name work email company role

What matters most to your agent? 0 of 3

What do you use for this today?

Which call would you make first?

What would your agent do with it?