Question 1

Can I run Gemma hosted or locally?

Accepted Answer

Both. The hosted path runs through Ollama Cloud with no hardware to provision. The local path runs on a machine you control. Because it is the same model family either way, a proof of concept on hosted infrastructure can become a self-hosted deployment without re-evaluating a different model.

Question 2

Which should I start with?

Accepted Answer

Start hosted if you are still deciding whether the answers hold up on your own questions, since there is nothing to stand up. Move local when data residency requires it, or when query volume makes hosted pricing the larger cost. The tiers usually differ in model size, so test the local one against real questions before committing to it.

Question 3

Does my data leave my network?

Accepted Answer

Running locally, no. Prompts and the operational data behind them stay inside your network. Running hosted, requests go to Ollama, which is a hosted service and should be evaluated as one. The deployment choice decides this.

Question 4

How does Gemma compare to a frontier model?

Accepted Answer

It trades some depth for control and cost. For routine questions about orders, shipments, and invoices it holds up, and it runs where a frontier API cannot. For the hardest multi-step analysis, routing can still send that work to a frontier provider while everything else stays on Gemma.

Question 5

How does Gemma differ from Gemini?

Accepted Answer

Same Google lineage, different deployment. Gemini is the hosted frontier family reached through Google's API. Gemma is open-weight, so it runs through Ollama Cloud or on your own hardware, and nothing leaves your network in the local case.

Question 6

Which other models can I run alongside Gemma?

Accepted Answer

Any of them, and more than one at a time. CommerceShip runs ChatGPT, Claude, Gemini, Mistral, and Muse as frontier models, and DeepSeek, Gemma, GLM, Kimi, Llama, MiniMax, Nemotron, and Qwen as open-weight models, served through Ollama Cloud or on your own hardware. Route routine work to one model and heavier analysis to another, and if a provider is rate-limited or down, the run moves to a fallback and completes.

Question 7

What access does CommerceShip need to Gemma?

Accepted Answer

The connection uses scoped credentials limited to the record types the integration works with, and you choose which fields are included. Credentials are encrypted at rest and stored separately from your operational data. Revoke the connection from either side at any time and the access ends with it.

Question 8

How is my data protected?

Accepted Answer

TLS in transit and AES-256 at rest, with tenant isolation enforced by row-level security in the database. Role-based permissions control who on your team can see cost data or change carrier settings, and automated actions are written to an audit log. CommerceShip is pursuing SOC 2 Type II certification. The security page carries the full detail.

Run Gemma hosted or on your own hardware

Gemma Integration Capabilities

How the Gemma integration works

Choose hosted or local

Assign Gemma to the work

Ask and act

Gemma Integration FAQs

Start hosted, move to your own hardware later

Other Models