feat(byok): Added Ollama CLoud as BYOK provider (#923)

* feat: add Ollama Cloud to KNOWN_PROVIDERS as OpenAI-compatible BYOK provider

* feat: add ollama.com to isOpenAICompatible base URL detection

* feat: add Ollama Cloud models to SUGGESTED_MODELS_BY_PROTOCOL fallback list

* fix: use full Ollama Cloud model list from /api/tags, drop -cloud suffix

* feat: add Ollama Cloud as native protocol with NDJSON streaming and connection test support

* fix: remove ollama.com from OpenAI compatibility check

* feat: add token overrides for Ollama Cloud models to prevent truncation

* fix: extend inferApiProtocol and legacy migration to recognize ollama.com base URLs

* fix: normalize Ollama Cloud base URL by stripping /api suffix during migration and in daemon

---------

Co-authored-by: herediaron <aronheredi346@gmail.com>
This commit is contained in:
Herédi Áron
2026-05-09 11:21:16 +08:00
committed by GitHub
co-authored by herediaron
parent 64ed8e7046
commit 66f84972cf
11 changed files with 389 additions and 6 deletions
+60
View File
@@ -190,6 +190,53 @@ export const KNOWN_PROVIDERS: KnownProvider[] = [
model: 'mimo-v2.5-pro',
models: ['mimo-v2.5-pro'],
},
{
label: 'Ollama Cloud',
protocol: 'ollama',
baseUrl: 'https://ollama.com',
model: 'gpt-oss:120b',
models: [
'cogito-2.1:671b',
'deepseek-v3.1:671b',
'deepseek-v3.2',
'deepseek-v4-flash',
'deepseek-v4-pro',
'devstral-2:123b',
'devstral-small-2:24b',
'gemini-3-flash-preview',
'gemma3:4b',
'gemma3:12b',
'gemma3:27b',
'gemma4:31b',
'glm-4.6',
'glm-4.7',
'glm-5',
'glm-5.1',
'gpt-oss:20b',
'gpt-oss:120b',
'kimi-k2:1t',
'kimi-k2-thinking',
'kimi-k2.5',
'kimi-k2.6',
'minimax-m2',
'minimax-m2.1',
'minimax-m2.5',
'minimax-m2.7',
'ministral-3:3b',
'ministral-3:8b',
'ministral-3:14b',
'mistral-large-3:675b',
'nemotron-3-nano:30b',
'nemotron-3-super',
'qwen3-coder:480b',
'qwen3-coder-next',
'qwen3-next:80b',
'qwen3-vl:235b',
'qwen3-vl:235b-instruct',
'qwen3.5:397b',
'rnj-1:8b',
],
},
{
label: 'MiMo (Xiaomi) — Anthropic',
protocol: 'anthropic',
@@ -233,6 +280,11 @@ function isValidOrbitTime(time: string): boolean {
function inferApiProtocol(model: string, baseUrl: string): ApiProtocol {
try {
const normalized = (baseUrl || '').toLowerCase();
// Any config pointing at ollama.com should resolve to the new ollama
// protocol so both chat and the connection test hit the native Ollama
// proxy instead of the Anthropic or OpenAI paths.
if (normalized.includes('ollama.com')) return 'ollama';
return isOpenAICompatible(model, baseUrl) ? 'openai' : 'anthropic';
} catch {
// Preserve the rest of the user's settings even if an old saved base URL is
@@ -286,6 +338,14 @@ export function loadConfig(): AppConfig {
// legacy config can be migrated when it is loaded.
if (!parsedHasApiProtocol) {
merged.apiProtocol = inferApiProtocol(merged.model, merged.baseUrl);
// Ollama Cloud legacy configs may carry a base URL that includes
// /api or /api/ — normalize to the host root so the daemon's own
// /api/chat appending doesn't double up.
if (merged.apiProtocol === 'ollama') {
merged.baseUrl = merged.baseUrl
.replace(/\/api\/?$/, '')
.replace(/\/+$/, '');
}
// Also set apiProviderBaseUrl so setApiProtocol() can correctly identify
// whether the user is on a known provider and switch defaults appropriately.
// null means "custom/unknown provider" so the protocol switch won't override
+42
View File
@@ -31,6 +31,48 @@ const OVERRIDES: Record<string, number> = {
// Spec: https://platform.deepseek.com/docs/model-cards
'deepseek-v4-pro': 384000,
'deepseek-v4-flash': 384000,
// Ollama Cloud models. LiteLLM keys this set under `ollama/`-prefixed
// ids (many with `-cloud` suffixes), so the bare model-id lookups never
// match. Add overrides so chat doesn't silently clip at 8192 tokens.
// 131072 (128k) is a safe floor for all Ollama Cloud models.
'cogito-2.1:671b': 131072,
'deepseek-v3.1:671b': 163840,
'deepseek-v3.2': 163840,
'devstral-2:123b': 131072,
'devstral-small-2:24b': 131072,
'gemini-3-flash-preview': 131072,
'gemma3:4b': 131072,
'gemma3:12b': 131072,
'gemma3:27b': 131072,
'gemma4:31b': 131072,
'glm-4.6': 131072,
'glm-4.7': 131072,
'glm-5': 131072,
'glm-5.1': 131072,
'gpt-oss:20b': 131072,
'gpt-oss:120b': 131072,
'kimi-k2:1t': 131072,
'kimi-k2-thinking': 131072,
'kimi-k2.5': 131072,
'kimi-k2.6': 131072,
'minimax-m2': 131072,
'minimax-m2.1': 131072,
'minimax-m2.5': 131072,
'minimax-m2.7': 131072,
'ministral-3:3b': 131072,
'ministral-3:8b': 131072,
'ministral-3:14b': 131072,
'mistral-large-3:675b': 131072,
'nemotron-3-nano:30b': 131072,
'nemotron-3-super': 131072,
'qwen3-coder:480b': 262144,
'qwen3-coder-next': 131072,
'qwen3-next:80b': 131072,
'qwen3-vl:235b': 131072,
'qwen3-vl:235b-instruct': 131072,
'qwen3.5:397b': 131072,
'rnj-1:8b': 131072,
};
export function modelMaxTokensDefault(model: string): number {