AI-built analytics · runs where your data lives · 52.23°N 21.01°E
The report that took two days
now takes ten minutes.
Daedalus is an AI assistant that builds working analytics. Tell it what you need in plain English — the weekly scorecard, the store pack, the customer list — and it assembles the whole workflow for you. It fits the systems you already have: it reads your files and databases where they live and never moves them. And once built, the workflow is yours — scheduled, free to re-run forever, with the finished report delivered wherever you expect it.
- Fits the files & systems you already have
- Your data never leaves the building
- Scheduling included — reports arrive on their own
- Maps, drive-times & spatial joins built in
The same report. Rebuilt by hand.
Every single week.
Someone on your team spends days every month pulling exports out of spreadsheets, ERP and TMS systems, patching them together, and formatting the same PDF — again. The tools that could automate it price for enterprises, sell scheduling as an add-on, or demand the data move to their cloud first. Daedalus does the whole job inside the setup you already have — AI builds the workflow, location analysis included, it runs on a schedule, and you own it — free to re-run.
Geospatial at the core · 40.71°N 74.00°W
The questions that start with "where"
Most tools stop at charts over rows and columns — but the questions that move a business usually depend on where: which sites, which customers, which zones, how far by road. Daedalus makes spatial analysis a first-class part of every workflow, and the AI builds it from the same plain-English prompt.
Trade areas & drive-time catchment
Who's within 5 miles — or 15 minutes' drive — of each store, how catchments overlap, and how many customers each location really owns.
Proximity & nearest
The nearest store, facility, or competitor to every point — along the road network, not just as the crow flies — and what's reachable from where.
Zones, territories & density
Which sites sit in which flood zone, sales territory, or service area — point-in-polygon joins across millions of rows, density on a hex grid.
Start from street addresses
No coordinates? Geocode plain addresses into points — on your machine, offline — then run any spatial step on top.
Site selection & risk
Score candidate locations by what surrounds them, and flag the assets that fall inside a risk or regulatory boundary.
Huge spatial files, read in place
GeoParquet, Parquet, CSV, and PostGIS are queried where they live — the worst case for "upload it first," handled without moving a byte.
Logistics & fleets · 52.29°N 21.05°E
Forty vans, two thousand stops, one dispatcher
A morning dispatch is a stack of location problems: which orders can go out today, from which depot, in what order. Daedalus plans it with real road travel times, computed on your machine — and tomorrow morning the same workflow runs again on tomorrow's orders.
- Draw a 30-minute drive-time around each depot and see which orders make the same-day cut
- Assign every order to a depot by road travel time, not straight-line distance
- Sequence each van's stops so it drives the short way round
- Put a real door-to-door ETA on the customer's confirmation email
Routing works offline, from locally built road-network tiles — no per-request API fees, no rate limits.
From a sentence to a shippable workflow
-
1
Describe your goal
Type what you want in plain English — "find every customer within 5 miles of a store and rank stores by catchment." No setup, no code.
-
2
Review the plan
The assistant proposes an ordered, step-by-step plan. Refine any step or regenerate it — before a single thing runs.
-
3
Build & run
It builds each step — including spatial joins and distances — and runs them on your real data where it lives, with a live preview at every stage.
-
4
Reuse, scale & ship
You own a working workflow. Put it on a schedule, run it over a whole folder of files — and the finished report lands where you expect it: your inbox, your Drive, your systems. It never calls the AI again.
Keep the machine, not just the answer
Most AI hands you a result you can't reuse. Daedalus hands you the workflow that produced it — open it, change any step, put it on a schedule, and it runs again tomorrow on new files without you touching it. Because the AI only builds the workflow and never runs it, there is nothing left to hallucinate: every run is deterministic — the same data gives the same answer, every time.
From a sentence to a working workflow
- Describe your goal in plain English — no code, no setup
- The assistant proposes a plan you can refine before anything runs
- It builds and runs every step on your real data, with live previews
- Need a custom step? It writes that too
Yours to review, reshape, and re-run
- Every step is open and readable — a real workflow, not a black box
- Adjust any step by hand, or regenerate just the one you don't like
- Re-run it as often as you want — zero AI tokens
- Run it as an agent over many files and ship results automatically
| Capability | AI chat & agents | Traditional ETL | Daedalus |
|---|---|---|---|
| Spatial / location analysis | No real geometry | A separate GIS tool | Native — maps, joins, drive-times, zones |
| Speed to build | Fast | Slow, all by hand | Fast — AI builds it |
| Cost to run it again | Pays the model every time | Free | Free — zero tokens |
| See & trust each step | Black box | Yes | Yes — every step open |
| Repeatable & deterministic | No | Yes | Yes |
| Runs unattended, at scale | Fragile | Yes | Yes — over a whole folder |
| Your data stays put | Usually uploaded | Depends | Always — nothing leaves |
Read every line before it runs
Ask for a step that doesn't exist — "flag parcels that intersect the proposed flood zone" — and the assistant writes it as ordinary, readable code that lands right in your workflow. There's no hidden endpoint to trust; the step is the code you're looking at.
- Read it before it ever runs
- Edit it like any other step
- Re-runs use this saved step — never the AI again
- Bring your own AI key — Anthropic, OpenAI, Google, or AWS Bedrock
Only a small schema sample is sent to build a step — never your dataset.
import duckdb
from daedalus import node, Port
@node(
inputs=[
Port("input_0", type="Table"), # parcels
Port("input_1", type="Table"), # flood zones
],
outputs=[Port("result", type="Table")],
)
def transform(context) -> duckdb.DuckDBPyRelation:
# Plain Python. Yours to read, edit, and re-run for free.
parcels = context["input_0"]
zones = context["input_1"]
return parcels.join(
zones, "ST_Intersects(parcels.geom, zones.geom)"
)Built to pass a security review
Your data never leaves
Workflows run next to your data — on your laptop or anywhere on your network. Nothing is uploaded, so there's no upload step to secure or to explain to compliance.
Predictable AI spend
You pay to build a workflow once, not every time it runs. No runaway token bills from an agent looping in the background — the cost stops when the building does.
Audit-ready by design
Every step, including the AI-written ones, is open and reviewable. Approve a workflow the way you'd approve any other document, before it touches production.
No lock-in
A workflow is a self-contained file you own — graph, steps, and results in one folder. Move it, share it, keep it. Nothing is trapped in a vendor's cloud.
For your IT & security team · 51.50°N 0.12°W
The technical answer
Got this forwarded by a colleague? Here's what your platform and security reviewers will want to know — in their language.
Does our data leave the network?
No. Lightweight workers run next to each source and connect outbound to the desktop over WebSocket — only result sets come back. No inbound ports are opened on the data machine, and there's no broker or cloud relay in between.
Does it need internet to run?
No. Building and running pipelines is fully offline. The hosted service is reached only for sign-in and optional AI step-generation — never to move or store your data.
Where do AI keys live, and what does the model see?
You bring your own key — Anthropic, OpenAI, Google, or AWS Bedrock. Keys are encrypted at rest with per-user key derivation, and only a small schema sample — column names and types — is sent at generation time. Never your rows.
Can we audit what actually runs?
Yes. A project is a self-contained folder of plain Python and JSON you can review in a pull request, like any other code. Secrets stay out of the manifest, in local project secrets.
How does it handle large datasets?
Pure-SQL steps are fused into a single query and pushed down to the source, where the data already lives. Filtering 3% of a 1 GB file moves about 30 MB across the wire — not the whole file.
What telemetry is collected?
Crash and performance reporting is opt-in and off by default. When enabled, source data, secrets, and home-folder paths are stripped or redacted; left unconfigured, it's disabled entirely.
Want the full architecture write-up and a security-review pack? We'll share both when we launch.
We started Daedalus after watching one too many location questions die in a queue — the data was sitting right there, but the one GIS seat was booked and the cloud upload was never going to clear compliance. The map shouldn't be a separate department.
— Paweł, founder
Know when it launches
The first release is around the corner. Leave your email and we'll tell you the moment it's ready.