Skip to main content
How It Works

From Spatial Context to Verified Workflows

Axis connects maps, data, scripts, notebooks, models, runtimes, and acceptance criteria so spatial agents can run the work, verify the result, repair visible failures, and package useful runs for reuse.

ArcGIS
Excel
QGIS
FME
Python
MATLAB
Shapefile
GeoTIFF
CSV
Databricks
AWS
GCP
Azure
01

CONNECT

Bring the spatial context

Bring the project context your team already has: maps, data, scripts, notebooks, GIS projects, APIs, models, sample outputs, and acceptance criteria.

  • Connect folders, scripts, notebooks, and GIS projects
  • Capture AOIs, date windows, schemas, CRS, and output targets
  • Attach acceptance criteria and prior outputs
  • Keep migration inputs as one proof lane, not the whole product

Axis starts from the context your spatial team already trusts.

File Scanner
Python scripts (.py)12 files
QGIS projects (.qgz)3 files
Excel workbooks (.xlsx)8 files
47 files mapped, 23 connections found
02

STRUCTURE

Turn scattered inputs into agent-ready work

Axis structures the work before an agent runs it. The system identifies assumptions, missing questions, dependencies, runtime constraints, and proof criteria.

  • Extract dependencies between data, code, maps, and outputs
  • Ask targeted questions only where context is missing
  • Turn analyst judgement into reusable context
  • Make the planned workflow visible before execution

The agent should know what it is proving before it starts.

Interview
12 of 18

Why is the buffer set to 500m?

Regulatory requirement
Species dispersal range
Legacy default
03

PLAN

Choose the execution path

The agent plans how the work should run and where it belongs: Python, PostGIS, QGIS/PyQGIS, Earth Engine, Databricks, Snowflake, AWS, GCP, private APIs, or customer runtimes.

  • Select the right runtime for the task
  • Separate proof runs from production placement
  • Show the workflow graph and expected outputs
  • Keep humans in the loop where judgement matters

Axis owns the work contract, not every compute engine.

Workflow Review
Input: flood_zones.shp
Buffer (500m) → Clip → Reproject
Output: analysis.gpkg
Ready for code generation
04

RUN

Execute the spatial job

Agents run the work through the selected path: analysis, model inference, workflow reconstruction, data conversion, map production, or pipeline execution.

  • Use deterministic tools where they are better than generation
  • Generate or adapt code only when needed
  • Record tool attempts, runtime choices, and outputs
  • Keep migration generation as a specialised proof path

The run is traceable, not a hidden chatbot answer.

Code Generation
Notebooks generated23 files
Libraries usedGeoPandas, Rasterio
Audit rounds3 (all passed)
Auditor: PASS - no ArcGIS license required
05

VERIFY

Check the result against spatial proof

Outputs are checked against spatial, technical, and domain criteria before they are treated as useful work.

  • Validate CRS, geometry, schema, coverage, and lineage
  • Compare against legacy outputs or sample targets where available
  • Attach map evidence, tables, traces, and QA notes
  • Flag exactly what diverged when evidence does not match

Plausible is not enough. Spatial work needs proof.

Deployment failed - geometry error
Fix found - testing in isolation...
Redeployed successfully
06

REUSE

Package what worked

Useful runs can become reusable workflows, capabilities, agent skills, private MCPs, API-facing operations, or runtime-bound jobs.

  • Save repeatable workflows with inputs, outputs, and evidence
  • Promote work into the customer's preferred runtime or product surface
  • Preserve corrections and evals for future runs
  • Turn migration proof into reusable spatial capability memory

The goal is production-ready spatial work your team can rerun.

Output Verification
Row count match12,847 / 12,847
CRS matchEPSG:4326
Spatial accuracy99.97%

You pay for a finished room, not bricks and drawings

When a contractor builds you a room, you pay for a completed room with a structural certificate - not a pile of bricks and a blueprint. Axis Agents works the same way. You get a running, verified pipeline - not just generated code that might work.

Before & After

Desktop ArcPy vs Automated Pipeline

Comparison of Desktop ArcPy vs Automated Pipeline capabilities
AspectDesktop ArcPyAutomated Pipeline
ConcurrencySingle-threadedDistributed (Spark)
Cost ModelPer-seat licenseCompute-based
EnvironmentLocal workstationCloud cluster
MonitoringManual checksBuilt-in observability
Version ControlFile timestampsGit-based CI/CD
Error HandlingManual debuggingAutonomous correction
Output ValidationManual spot checksAutomated spatial comparison
Built for Enterprise

Why GIS Teams Trust This Approach

Context Before Automation

Axis starts by absorbing the maps, data, scripts, notebooks, outputs, requirements, and constraints that already define the work.

Multi-Tool Spatial Understanding

Agents reason across GIS tools, Python, notebooks, spatial formats, APIs, runtimes, and domain-specific acceptance criteria.

Human Judgement Captured

Analyst choices, edge cases, missing questions, and review decisions become reusable context instead of staying in one person's head.

Proof And Correction Loops

Failures are diagnosed and recorded. Outputs need evidence before a run becomes a workflow, capability, or agent.

Common Questions

Frequently Asked Questions

Does my code or data leave my infrastructure?

No. The scanner runs locally on your machine. Only workflow metadata (file names, detected operations) is sent to AI for analysis. Your actual geodata files never leave your environment. Generated code deploys to your own cloud account.

How long does the process take?

Scanning takes minutes. The interview phase depends on your team's availability - typically 1-3 days. Code generation takes hours. End-to-end, most workflows are automated within 1-2 weeks.

What if the generated code is wrong?

First, an Auditor agent reviews every line before you see it. If issues slip through and deployment fails, the system automatically diagnoses the error, searches a knowledge base of known fixes, tests the fix in isolation, and redeploys. This loop runs up to three times. Then your team reviews. Finally, the system compares cloud output against your legacy results to verify accuracy.

Do I need ArcGIS licenses to run the output?

No. Generated pipelines use open Python libraries (GeoPandas, Rasterio, GDAL). Run on 100 machines without buying 100 ArcGIS licenses. Scale processing without per-seat costs.

How do I know the cloud output matches my legacy system?

The Verify stage automatically compares cloud results against your legacy output - row counts, coordinate reference systems, and spatial accuracy. Your pipeline is only marked as complete when outputs match. Any discrepancies are flagged with exactly what diverged.

Compatibility

What We Support

Input Formats

  • ArcPy scripts (.py)
  • QGIS projects (.qgz, .qgs)
  • Excel with formulas (.xlsx, .xlsm)
  • FME workbenches (.fmw)
  • ModelBuilder exports
  • Shapefiles, geodatabases, GeoPackage

Output Formats

  • Cloud Optimised GeoTIFF (COG)
  • GeoParquet
  • STAC Catalogues
  • PMTiles
  • FlatGeobuf
  • Delta Lake tables
Start With One Workflow

Pick your messiest process. See results in days.

No 6-month proposal. No massive commitment.

NDA protection for all discussions.