Ollama Cloud Provider Setup Guide

Ollama Cloud is the hosted model API that Ollama operates. The ollama-cloud provider talks to it straight at https://ollama.com using Ollama's native /api/chat API. No local Ollama server is involved, and no local Ollama app needs to be signed into cloud mode. Reference models with refs such as ollama-cloud/kimi-k2.6.

OpenClaw registers ollama-cloud under its own provider id, which keeps cloud-only credentials, live catalog discovery, and model selection separate from a local ollama host. For local Ollama, hybrid cloud-plus-local routing, embeddings, and custom host details, see Ollama.

Setup

Work through Ollama's API key instructions, then execute:

openclaw onboard --auth-choice ollama-cloud

Alternatively, set:

export OLLAMA_API_KEY="<your-ollama-cloud-api-key>" # pragma: allowlist secret

When onboarding runs non-interactively, the key can be passed straight in:

openclaw onboard --auth-choice ollama-cloud --ollama-cloud-api-key "<key>"

The default model chosen by onboarding is ollama-cloud/minimax-m2.7.

Defaults

Provider: ollama-cloud
Base URL: https://ollama.com
Env var: OLLAMA_API_KEY
API style: Ollama native /api/chat
Onboarding default model: ollama-cloud/minimax-m2.7

When to choose Ollama Cloud

You would rather use hosted Ollama models than run ollama serve on your own machine.
You want the native Ollama chat API shape that OpenClaw already uses for local Ollama, except aimed at https://ollama.com.
You need a straightforward cloud route to models that Ollama already hosts in its catalog.
You have no need for local model pulls, local GPU control, or inference confined to a LAN.

Pick Ollama instead if you want local-only routing, or cloud-plus-local routing through a signed-in Ollama host. Pick an OpenAI-compatible provider instead when /v1/chat/completions semantics or OpenAI-style, provider-specific features are what you need.

Models

An API key is mandatory for this provider; with no key it remains inactive. Once a key is present, OpenClaw pulls Ollama Cloud models live from the hosted catalog:

openclaw models list --provider ollama-cloud
openclaw models set ollama-cloud/kimi-k2.6

The live catalog's hosted ids cover deepseek-v4-flash, glm-5.2, gpt-oss:20b, kimi-k3, and minimax-m3. If discovery fails, the most recent successful inventory for the same credentials is retained. When no earlier inventory exists, OpenClaw falls back to bundled suggestions and logs the discovery failure. A successful response that returns nothing wipes the discovered models, and any later failure leaves that empty result in place. The retired kimi-k2.5 is still flagged deprecated for exact references that already exist, though it is no longer a hosted model in the current catalog.

Cloud catalog ids are what model ids refer to here, not local pull names. Should a model name resolve on a local Ollama host yet be missing from the hosted catalog, switch to the ollama provider with that local host.

Live test

To smoke-test an Ollama Cloud API key, aim the Ollama live test at the hosted endpoint and pick a model from your current catalog:

export OLLAMA_API_KEY="<your-ollama-cloud-api-key>" # pragma: allowlist secret

OPENCLAW_LIVE_TEST=1 \
OPENCLAW_LIVE_OLLAMA=1 \
OPENCLAW_LIVE_OLLAMA_BASE_URL=https://ollama.com \
OPENCLAW_LIVE_OLLAMA_MODEL=kimi-k2.6 \
pnpm test:live -- extensions/ollama/ollama.live.test.ts

The cloud smoke covers text, native stream, and web search; pass OPENCLAW_LIVE_OLLAMA_WEB_SEARCH=0 to leave web search out. Embeddings are skipped by default for https://ollama.com, since Ollama Cloud API keys might not authorize /api/embed; use OPENCLAW_LIVE_OLLAMA_EMBEDDINGS=1 to force them.

Troubleshooting

Ollama Cloud requires an API key / Set OLLAMA_API_KEY errors: supply a genuine cloud API key. The local ollama-local marker applies only to local or private Ollama hosts.
Unknown model errors: execute openclaw models list --provider ollama-cloud and copy the hosted model id verbatim.
Tool-call or raw JSON problems on custom Ollama hosts: verify you are not unintentionally hitting an OpenAI-compatible /v1 URL. Ollama routes need the native base URL, with no /v1 suffix.