Ollama Cloud (hosted inference)

Fact-checked 2026-06-27 against https://docs.ollama.com/cloud and https://ollama.com/blog/cloud-models; cross-referenced with https://fabiorehm.com/blog/2026/04/12/pi-ollama-cloud-api/ (hands-on April 2026).

Distinct from local Ollama (docs/providers/ollama.md): local Ollama runs at http://localhost:11434/v1 with no API key; Ollama Cloud is a hosted service at https://ollama.com/v1 that requires OLLAMA_API_KEY. They share an underlying Ollama engine but are independent providers in Cyberine — do NOT conflate the two or reuse the keyless local path for cloud.

Endpoint & auth


OpenAI-compat base URL	`https://ollama.com/v1`
Native base	`https://ollama.com/api`
Auth header	`Authorization: Bearer <OLLAMA_CLOUD_API_KEY>`
Keys	https://ollama.com/settings/keys (free tier available)

The /api prefix serves native Ollama endpoints (/api/chat, /api/tags). The OpenAI-compatible layer lives at /v1 — a common setup mistake is including /api in the base URL. Confirmed by multiple independent sources including the official docs and hands-on testing (fabiorehm.com/blog April 2026 explicitly notes the /api prefix pitfall).

Endpoints

Method + path	Purpose
`POST /v1/chat/completions`	OpenAI-compat chat — what `LoopAgentAdapter` calls
`GET /v1/models`	List cloud models available to this key
`GET /api/tags`	Native model list (cloud models appear with `:cloud` suffix)

Cloud models (as of 2026-06-27)

Source: https://ollama.com/blog/cloud-models (official Ollama blog announcement)

Model id	Notes
`qwen3-coder:480b-cloud`	Qwen3 Coder 480B — large code-focused model
`gpt-oss:120b-cloud`	120B parameter open-weights model
`gpt-oss:20b-cloud`	20B parameter open-weights model
`deepseek-v3.1:671b-cloud`	DeepSeek V3.1 671B

Additional models observed in community testing (UNVERIFIED against official catalog — may vary by plan tier or may have been added after the blog post): kimi-k2.6, glm-4.6, glm-5:cloud, minimax-m3, qwen3-next:80b, cogito-2.1:671b. Retrieve the live list for your key with:

curl https://ollama.com/v1/models -H "Authorization: Bearer $OLLAMA_CLOUD_API_KEY"

The :cloud suffix in a model id signals that inference runs on Ollama's datacenter hardware, not locally. The same model without the suffix pulls and runs locally (standard local Ollama behaviour).

Local vs Cloud distinction

	Local Ollama	Ollama Cloud
Base URL	`http://localhost:11434/v1`	`https://ollama.com/v1`
API key	None (static `'ollama'` placeholder)	Required (`OLLAMA_API_KEY`)
Model source	Locally pulled (`ollama pull <model>`)	Hosted catalog; `:cloud`-suffix ids
GPU required	Yes (host hardware)	No (Ollama's datacenter)
Cyberine env	`OLLAMA_BASE_URL` / `OLLAMA_MODEL`	`OLLAMA_CLOUD_API_KEY` / `OLLAMA_CLOUD_BASE_URL` / `OLLAMA_CLOUD_MODEL`
Cyberine flag	`ollamaDirect`	`ollamaCloudDirect`

Cyberine wiring

Env: OLLAMA_CLOUD_API_KEY (enables), OLLAMA_CLOUD_BASE_URL (default https://ollama.com/v1), OLLAMA_CLOUD_MODEL.
Registry id: ollama-cloud (cyberine auth login ollama-cloud, chat provider picker).
Availability flag: ollamaCloudDirect (key present).
Compare-fleet adapter: loop-ollama-cloud-direct (read-only tools) — wired in buildAdapters.ts.
Build worker: loop-ollama-cloud-build — opt-in via CYBERINE_BUILD_FLEET=loop-ollama-cloud-build.

Sources

https://docs.ollama.com/cloud (official Ollama Cloud documentation)
https://ollama.com/blog/cloud-models (official cloud-models announcement; verified model ids)
https://fabiorehm.com/blog/2026/04/12/pi-ollama-cloud-api/ (hands-on setup, April 2026; confirms https://ollama.com/v1 base URL and /v1/models endpoint with auth)