Where it runs, and what it reaches.
Where you put the server sets your jurisdiction. How many systems it reaches sets whether the answer is complete. Both get settled before a scan starts.
Three deployment models
| Your own servers | Cloud region you control | Hosted service | |
|---|---|---|---|
| Index location | Your infrastructure | Your tenancy, region of your choosing | Operated for you |
| Who runs the server | Your team | Your team | The vendor |
| Jurisdiction control | Complete | Complete | Confirmed by contract |
| Outbound connection | Not required for discovery | Within your cloud account | Standard hosted access |
| Typical fit | Government, healthcare, defence work | Enterprises already standardized on a cloud | Teams that want capability without operating a server |
Jurisdiction is a deployment choice
GEODI runs inside Canada, the United States or Mexico depending on where you put the server. Data residency stops being a vendor promise and becomes a fact about your own topology.
Two processing paths deserve naming, since they are the only ones a privacy reviewer needs to interrogate.
- Optical character recognition
- Runs on the GEODI server. Scanned pages are read locally and no page image is transmitted anywhere.
- AI question answering
- Uses retrieval augmented generation. The prompt and retrieved context reach the model, source documents do not. Where policy forbids any external inference call, the model can be self hosted so nothing leaves.
- Index and classification
- Local to the server in every deployment model.
- Remediation
- Executed at the source, under credentials you supply and can revoke.
Source coverage
More than 200 adapters in total. Content is processed identically whatever container it arrives in, so a spreadsheet on a share, the same file attached to an email, and a record inside a database are treated the same way.
Files and storage
- Folder and network share
- FTP
- SSH
- WebDAV
- Amazon S3
Cloud drives
- Dropbox
- Google Drive
- One Drive
Collaboration
- SharePoint
- Microsoft Teams
- Confluence
- Jira
- Exchange and mail platforms
- Attachments read as content
Structured data
- Relational databases
- Sampling by record count
Other sources
- Web content
- Feed sources
- Endpoints through the agent
| Operation | Access needed |
|---|---|
| Discovery and indexing | Read only |
| Classification | Read only |
| Masking for display | Read only |
| Remediation | Enabled per source, separate credentials |
A scan holds no write access
Indexing runs read only. Remediation is a separate switch, turned on per source with credentials you scope and revoke independently. A misconfigured scan has no rights to change a file.
Every action that does change something writes a record naming the account, the target and the time.