Skip to main content
How It Works

From Scattered Files to Self-Correcting, Verified Pipelines

Axis Agents scans your existing workflow folder - ArcPy scripts, Excel files, QGIS projects - and reconstructs the actual process. AI agents interview your analysts, generate code, deploy it, fix errors automatically, and verify the output matches your legacy system.

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

DISCOVER

Static Analysis & Dependency Mapping

Point Axis Agents at your workflow folder. It identifies 30+ GIS file types, parses code to extract ArcPy operations, reads metadata, and maps how files connect to each other.

  • Detects ArcPy, QGIS, FME, and Excel files
  • Parses Python code using AST analysis
  • Extracts shapefile schemas and CRS metadata
  • Builds a dependency graph of your entire workflow

Runs locally. Your files never leave your machine.

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

REFINE

Logic Extraction & Knowledge Capture

AI generates questions based on what it could not figure out automatically. Why is the buffer 500 metres? Where does this data come from? Your analysts answer, capturing knowledge that exists only in their heads.

  • Questions generated from code analysis gaps
  • Multiple choice options to reduce typing
  • Delegate questions to colleagues who know the answer
  • Progress saves automatically - return anytime

Captures undocumented knowledge that survives staff turnover.

Interview
12 of 18

Why is the buffer set to 500m?

Regulatory requirement
Species dispersal range
Legacy default
03

REVIEW

Human-in-the-Loop Review

Before any code is written, you see the proposed workflow as a visual graph. Each processing step, each data dependency. Your team reviews, corrects mistakes, and approves the plan.

  • Visual DAG shows all processing steps
  • Edit nodes, add steps, change connections
  • See how interview answers affected the plan
  • Nothing happens without your sign-off

No black box. Full transparency before generation.

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

GENERATE

Multi-Agent Code Generation & Audit

AI agents write Python code using open libraries - GeoPandas, Rasterio, GDAL. A separate Auditor agent reviews every line. If it finds issues, the Builder revises automatically - up to three rounds until the code passes.

  • ArcPy translates to GeoPandas + Rasterio
  • Output formats: GeoParquet, COG, STAC catalogues
  • Auditor agent validates correctness automatically
  • Multi-round revision loop until quality passes

You own the code. No vendor lock-in. Run on unlimited machines.

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

DEPLOY & CORRECT

Self-Correcting Cloud Deployment

Deploy to Databricks, AWS, GCP, or Snowflake. If deployment fails, AI agents diagnose the error, search a knowledge base of thousands of documented GIS issues, test the fix in isolation, and redeploy. Up to three automatic attempts before escalating to your team.

  • One-click deployment to your cloud platform
  • Automatic error classification and diagnosis
  • Knowledge base of thousands of known GIS issues
  • Each fix tested in isolation before redeployment

Most tools stop at deployment. Axis Agents runs until it works.

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

VERIFY

Automated Output Comparison

Your pipeline is only complete when outputs match. The system compares cloud results against your legacy output - row counts, coordinate reference systems, spatial accuracy. If something doesn't match, it flags exactly what diverged.

  • Row count comparison between legacy and cloud
  • CRS matching and spatial accuracy checks
  • Automated reporting of any discrepancies
  • Pipeline marked complete only when verified

Not a report. A certificate that your migration is correct.

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

Your Data Stays Local

File scanning runs entirely on your machine. Only workflow metadata is sent for AI analysis - never your actual geodata files.

Multi-Tool Understanding

Handles workflows spanning Excel, ArcGIS, QGIS, FME, and Python. Detects where data flows between tools and preserves that logic.

Knowledge Capture

The interview system extracts tacit knowledge: why parameters were chosen, what happens in edge cases. This survives staff turnover.

Self-Correcting Deployment

If deployment fails, the system diagnoses the error, finds a known fix, tests it, and redeploys automatically. No manual debugging.

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.