奥拉玛云模型

Runs on Ollama infrastructure; requires sign-in, network access and available credits.

ollama run gemma4:31b-cloud "Explain Kubernetes admission control" ```

Local models require suitable hardware and a download before offline use. Memory needs vary with quantization, context, and concurrency; download size is not a complete RAM/VRAM requirement.

Getting started

Install Ollama using its official installer. Open the app or run ollama and follow setup prompts. Choose a local model, or sign in for cloud access. On Linux, start ollama serve if the server is not already running. Official quickstart.

```bash

Quick decision guide

Editorial starting points, not benchmark rankings. Evaluate representative prompts, latency, quality, and cost. Tool support means a model can request tools; your application must implement and authorize their execution.

Structured-output limitation: official documentation. Recommendations are maintained prose; regeneration changes only the marked comparison block.

API examples

These Bash examples use curl with connection and total timeouts. Run local inference after installing Ollama and downloading gemma4:e2b:

curl --fail-with-body --silent --show-error \
  --connect-timeout 10 --max-time 120 \
  http://localhost:11434/api/chat \
  -H 'Content-Type: application/json' \
  -d '{"model":"gemma4:e2b","messages":[{"role":"user","content":"Explain Kubernetes admission control"}],"stream":false}'

To route through your signed-in local server to cloud, replace the model with gemma4:31b-cloud. A cloud suffix changes where inference executes, even when the HTTP endpoint is localhost.

For direct cloud access, create an Ollama API key and supply it through your environment or secret manager. This reads it without echoing it or putting its value in shell history:

read -r -s -p 'Ollama API key: ' OLLAMA_API_KEY; printf '\n'
export OLLAMA_API_KEY
: "${OLLAMA_API_KEY:?Set OLLAMA_API_KEY first}"
curl --fail-with-body --silent --show-error \
  --connect-timeout 10 --max-time 120 \
  https://ollama.com/api/chat \
  -H "Authorization: Bearer ${OLLAMA_API_KEY}" \
  -H 'Content-Type: application/json' \
  -d '{"model":"gemma4:31b","messages":[{"role":"user","content":"Explain Kubernetes admission control"}],"stream":false}'

In PowerShell 7, use a secret-store-provided $env:OLLAMA_API_KEY:

if (-not $env:OLLAMA_API_KEY) { throw 'Set OLLAMA_API_KEY from your secret store first' }
$requestBody = @{
    model = 'gemma4:31b'
    messages = @(@{ role = 'user'; content = 'Explain Kubernetes admission control' })
    stream = $false
} | ConvertTo-Json -Depth 5
Invoke-RestMethod -Method Post -Uri 'https://ollama.com/api/chat' `
    -Headers @{ Authorization = "Bearer $env:OLLAMA_API_KEY" } `
    -ContentType 'application/json' -Body $requestBody -TimeoutSec 120

Public model discovery requires no API key:

curl --fail-with-body --silent --show-error \
  --connect-timeout 10 --max-time 30 https://ollama.com/api/tags

See API base URLs and authentication. Examples were checked against the docs; paid inference was not executed in this audit. Avoid verbose HTTP tracing with credentials.

Engineering workflows

Use these as review prompts with an appropriate model. Output is a draft to test, not authority to deploy:

Remove secrets and private data before sending material to cloud models. Keep agent credentials separate from prompts.

Integrations

The existing references remain useful. Configure each tool for the intended endpoint/model and check its own authentication and feature support:

Local versus cloud

Tag namespaces differ. For direct cloud requests, copy an identifier from https://ollama.com/api/tags. Use the catalog's separately verified CLI tag through local Ollama. Do not mechanically strip :cloud: aliases and dated identifiers can differ. For example, the observed direct ID is deepseek-v4-pro:0813.

Ollama says cloud prompts/responses are not used for training; see its data-handling documentation and privacy policy. Cloud requests still leave your machine. For local-only operation, the FAQ documents OLLAMA_NO_CLOUD=1; restart Ollama after changing it. Keep the local server on loopback unless you add appropriate network access controls.

Free versus paid

The live Ollama pricing page, checked October 14, 2026, advertises token-priced usage credits:

Free starter usage applies to a smaller starter set; purchased credits unlock all models. Included credits reset monthly and do not roll over. The page says switching from an existing Pro/Max plan removes the old session/weekly limits; existing subscribers should check their account terms. Queuing is bounded. We do not infer entitlement from a model's relative usage band. Enterprise is custom-priced.

Charges depend on input, cached-input and output tokens. Consult the generated price table and current billing page before spending. Off-peak DeepSeek rates are documented per entry. Local weights do not consume cloud credits.