Deployment
On-premise process discovery
The same current-state map, and the files never leave your machine. ThinkPrior does not receive the documents. Model calls go only to the models you configure.
Appliance
- Source · Local volumeClient SOPs
- Source · Local volumeTranscripts
- Source · Local volumeRecordings
- Boundary · Does not receive filesThinkPrior
- Action Step · Five slots you configureYour models
Engagement files stay on the appliance. ThinkPrior does not receive them.
01 · The same map, on a machine you run
What is on-premise process discovery?
On-premise process discovery is the same document-to-map workflow, installed as an appliance you run.
You request the appliance. There is no public download. The engagement still starts from SOPs, transcripts, policies, decks, and recordings, and it still produces a current-state map with gaps left visible.
The difference is where the corpus lives. On the appliance, those files stay on local volumes. ThinkPrior as a vendor does not receive them.
Teams that cannot upload a client pile to a vendor cloud use this path.
No download
You request the appliance. There is no public installer.
Same map
Gaps stay visible. The workflow does not change.
02 · Documents stay on that machine
Where do the documents go?
Documents stay on the machine you run.
The appliance keeps the corpus, the maps, and the audit logs on local volumes, and it does not call ThinkPrior or Google Cloud for those documents. A cloud trial is a different deployment with its own protections, described on the data protection page. Choose the on-premise path when the files themselves must not leave the operator's machine.
Inside the appliance the work is unchanged: the analyst confirms scope, finishes the gaps, and exports the same map.
Local volumes
The corpus stays on the machine you run.
Audit logs
Maps and audit logs stay on local volumes too.
03 · Air-gapped when your models stay inside
Is the appliance air-gapped?
The appliance is air-gapped only when you point the models at services that never leave your network.
The appliance itself does not send engagement documents to ThinkPrior. Model calls go to the endpoints you configure. On-premise setup is self-serve across five slots: fast, reasoning, embeddings, vision, and speech.
Keys stay on your machine. Routing does not log keys, prompts, or documents. Cloud enterprise bring-your-own-key is provisioned separately.
It is not a toggle on the cloud trial.
Five slots
Fast, reasoning, embeddings, vision, and speech.
Air-gapped
Only when the models never leave your network.
04 · A desktop for a trial, a Linux server for production
What does the appliance need to run?
A trial runs on a Mac or Windows machine with Docker Desktop, 16 GB of RAM, and 50 GB of free disk.
A production engagement runs on an Ubuntu 22.04 or RHEL 9 server, with 8 vCPU and 32 GB of RAM recommended. No GPU is required, because model calls go to the endpoints you configure. A setup wizard checks Docker, picks the ports, deploys the services, and creates the first admin account.
You then add model keys in the five slots, or load a key file, and the appliance is ready for the first upload. Request the appliance from the trial form to receive the package.
16 GB
RAM for a desktop trial, with 50 GB of free disk.
32 GB
Recommended for a production Linux server, with 8 vCPU.
ThinkPrior, on your machine.
Same loop. Documents stay on your machine. ThinkPrior does not receive them.
- Runs on your machine
- Documents stay with you
- Your models
- ThinkPrior never receives your files
Your machine
Browser
Appliance
Your files
Your models
Browser
Appliance
Your files
Your models
Use your own models. Calls go only where you point them.