@oh-my-pi/pi-ai 14.5.0 → 14.5.2

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## [Unreleased]
 
+## [14.5.1] - 2026-04-26
+
+### Fixed
+
+- Fixed NVIDIA NIM DeepSeek-V4 models leaking chat-template tool-call markers (e.g. `<｜DSML｜tool_calls｜>`) into visible response text by stripping the special tokens from streamed `delta.content` ([#798](https://github.com/can1357/oh-my-pi/issues/798))
+
 ## [14.4.0] - 2026-04-26
 
 ### Added
@@ -89,6 +95,7 @@
 - Fixed shell execution failure responses to preserve all result fields when sanitizing, preventing truncated metadata in stream results
 - Fixed context overflow detection to recognize `model_context_window_exceeded` from z.ai / GLM providers, preventing infinite retry loops when context window is exceeded ([#638](https://github.com/can1357/oh-my-pi/issues/638))
 - Fixed strict tool schema enforcement to preserve `additionalProperties: false` and required keys for reused nested object schemas, preventing invalid `todo_write` function schemas in Codex/OpenAI requests
+- Fixed GitHub Copilot reasoning regressions by preserving GPT-5.x / Claude 4.x reasoning controls instead of stripping them from requests ([#773](https://github.com/can1357/oh-my-pi/issues/773))
 
 ## [14.1.0] - 2026-04-11
 

package/README.md

@@ -72,6 +72,7 @@ Unified LLM API with automatic model discovery, provider configuration, token an
 - **Qwen Portal** (supports `QWEN_OAUTH_TOKEN` or `QWEN_PORTAL_API_KEY`)
 - **Cloudflare AI Gateway** (requires `CLOUDFLARE_AI_GATEWAY_API_KEY` and provider-specific gateway base URL)
 - **Ollama** (local OpenAI-compatible runtime; optional `OLLAMA_API_KEY`)
+- **Ollama Cloud** (hosted native Ollama API; requires `OLLAMA_CLOUD_API_KEY`)
 - **llama.cpp** (local OpenAI and Anthropic compatible inference server)
 - **vLLM** (OpenAI-compatible server; `VLLM_API_KEY` for secured deployments)
 - **GitHub Copilot** (requires OAuth, see below)
@@ -690,13 +691,14 @@ console.log(`Using ${model.name} via ${model.api} API`);
 
 ### Custom Models
 
-You can create custom models for local inference servers or custom endpoints:
-For Ollama, `OLLAMA_API_KEY` is optional and mainly needed for authenticated/self-hosted gateways.
+You can create custom models for local inference servers or custom endpoints.
+
+For local Ollama, `OLLAMA_API_KEY` is optional and mainly needed for authenticated/self-hosted gateways. `ollama` remains the local OpenAI-compatible runtime integration.
 
 ```typescript
 import { Model, stream } from "@oh-my-pi/pi-ai";
 
-// Example: Ollama using OpenAI-compatible API
+// Example: local Ollama using the OpenAI-compatible API
 const ollamaModel: Model<"openai-completions"> = {
 	id: "llama-3.1-8b",
 	name: "Llama 3.1 8B (Ollama)",
@@ -710,6 +712,28 @@ const ollamaModel: Model<"openai-completions"> = {
 	maxTokens: 32000,
 };
 
+const localResponse = await stream(ollamaModel, context, {
+	apiKey: process.env.OLLAMA_API_KEY, // Optional; local Ollama usually runs without auth
+});
+
+// Example: Ollama Cloud using the native /api/chat transport
+const ollamaCloudModel: Model<"ollama-chat"> = {
+	id: "gpt-oss:120b",
+	name: "GPT OSS 120B (Ollama Cloud)",
+	api: "ollama-chat",
+	provider: "ollama-cloud",
+	baseUrl: "https://ollama.com",
+	reasoning: true,
+	input: ["text", "image"],
+	cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+	contextWindow: 262144,
+	maxTokens: 8192,
+};
+
+const cloudResponse = await stream(ollamaCloudModel, context, {
+	apiKey: process.env.OLLAMA_CLOUD_API_KEY,
+});
+
 // Example: LiteLLM proxy with explicit compat settings
 const litellmModel: Model<"openai-completions"> = {
 	id: "gpt-4o",
@@ -744,11 +768,6 @@ const proxyModel: Model<"anthropic-messages"> = {
 		"X-Custom-Auth": "bearer-token-here",
 	},
 };
-
-// Use the custom model
-const response = await stream(ollamaModel, context, {
-	apiKey: process.env.OLLAMA_API_KEY, // Optional; local Ollama usually runs without auth
-});
 ```
 
 ### OpenAI Compatibility Settings
@@ -928,6 +947,7 @@ In Node.js environments, you can set environment variables to avoid passing API
 | OpenRouter     | `OPENROUTER_API_KEY`                                                         |
 | LiteLLM        | `LITELLM_API_KEY`                                                            |
 | Ollama         | `OLLAMA_API_KEY` (optional for local deployments)                            |
+| Ollama Cloud   | `OLLAMA_CLOUD_API_KEY`                                                     |
 | Qwen Portal    | `QWEN_OAUTH_TOKEN` or `QWEN_PORTAL_API_KEY`                                  |
 | zAI            | `ZAI_API_KEY`                                                                |
 | MiniMax Code   | `MINIMAX_CODE_API_KEY` (international) or `MINIMAX_CODE_CN_API_KEY` (China) |
@@ -957,7 +977,8 @@ Provider endpoint defaults for the current OpenAI-compatible integrations:
 - ZenMux (OpenAI): `https://zenmux.ai/api/v1`
 - ZenMux (Anthropic models): `https://zenmux.ai/api/anthropic`
 - vLLM: `http://127.0.0.1:8000/v1`
-- Ollama: local OpenAI-compatible runtime
+- Ollama: local OpenAI-compatible runtime (`http://127.0.0.1:11434/v1`)
+- Ollama Cloud: native Ollama API host (`https://ollama.com/api`, configured here as base URL `https://ollama.com`)
 - LiteLLM: `http://localhost:4000/v1`
 - Cloudflare AI Gateway: `https://gateway.ai.cloudflare.com/v1/<account>/<gateway>/anthropic`
 - Qwen Portal: `https://portal.qwen.ai/v1`
@@ -1049,7 +1070,7 @@ Credentials are saved to `agent.db` in the agent directory. `/login qianfan` ope
 
 `login` supports OAuth providers (Anthropic, OpenAI Codex, GitHub Copilot, Gemini CLI, Antigravity) and API-key onboarding flows.
 
-For the current OpenAI-compatible integrations, API-key onboarding covers Together, Moonshot, Qianfan, NVIDIA, NanoGPT, Hugging Face, Venice, Xiaomi, vLLM, LiteLLM, Cloudflare AI Gateway, and Qwen Portal. Ollama is typically local and unauthenticated; set `OLLAMA_API_KEY` only when your Ollama deployment enforces bearer auth.
+For the current API-key onboarding flows, the library covers Together, Moonshot, Qianfan, NVIDIA, NanoGPT, Hugging Face, Venice, Xiaomi, vLLM, LiteLLM, Cloudflare AI Gateway, Qwen Portal, and Ollama Cloud. Ollama remains the local runtime integration; set `OLLAMA_API_KEY` only when your local or self-hosted deployment enforces bearer auth.
 
 ### Programmatic OAuth
 

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "14.5.0",
+	"version": "14.5.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -46,8 +46,8 @@
 		"@aws-sdk/credential-provider-node": "^3.972.36",
 		"@bufbuild/protobuf": "^2.12.0",
 		"@google/genai": "^1.50.1",
-		"@oh-my-pi/pi-natives": "14.5.0",
-		"@oh-my-pi/pi-utils": "14.5.0",
+		"@oh-my-pi/pi-natives": "14.5.2",
+		"@oh-my-pi/pi-utils": "14.5.2",
 		"@sinclair/typebox": "^0.34.49",
 		"@smithy/node-http-handler": "^4.6.1",
 		"ajv": "^8.20.0",
@@ -58,7 +58,7 @@
 		"zod": "4.3.6"
 	},
 	"devDependencies": {
-		"@types/bun": "^1.3.13"
+		"@types/bun": "^1.3"
 	},
 	"engines": {
 		"bun": ">=1.3.7"

package/src/api-registry.ts

@@ -24,6 +24,7 @@ const BUILTIN_APIS = new Set<KnownApi>([
 	"google-generative-ai",
 	"google-gemini-cli",
 	"google-vertex",
+	"ollama-chat",
 	"cursor-agent",
 ]);
 

package/src/auth-storage.ts

@@ -51,6 +51,7 @@ import { loginMoonshot } from "./utils/oauth/moonshot";
 import { loginNanoGPT } from "./utils/oauth/nanogpt";
 import { loginNvidia } from "./utils/oauth/nvidia";
 import { loginOllama } from "./utils/oauth/ollama";
+import { loginOllamaCloud } from "./utils/oauth/ollama-cloud";
 import { loginOpenAICodex } from "./utils/oauth/openai-codex";
 import { loginOpenCode } from "./utils/oauth/opencode";
 import { loginParallel } from "./utils/oauth/parallel";
@@ -838,6 +839,11 @@ export class AuthStorage {
 				await saveApiKeyCredential(apiKey);
 				return;
 			}
+			case "ollama-cloud": {
+				const apiKey = await loginOllamaCloud(ctrl);
+				await saveApiKeyCredential(apiKey);
+				return;
+			}
 			case "cerebras": {
 				const apiKey = await loginCerebras(ctrl);
 				await saveApiKeyCredential(apiKey);

package/src/cli.ts

@@ -12,6 +12,7 @@ import { loginKilo } from "./utils/oauth/kilo";
 import { loginKimi } from "./utils/oauth/kimi";
 import { loginMiniMaxCode, loginMiniMaxCodeCn } from "./utils/oauth/minimax-code";
 import { loginNanoGPT } from "./utils/oauth/nanogpt";
+import { loginOllamaCloud } from "./utils/oauth/ollama-cloud";
 import { loginOpenAICodex } from "./utils/oauth/openai-codex";
 import { loginParallel } from "./utils/oauth/parallel";
 import { loginTavily } from "./utils/oauth/tavily";
@@ -271,6 +272,23 @@ async function login(provider: OAuthProvider): Promise<void> {
 				console.log(`\nAPI key saved to ~/.omp/agent/agent.db`);
 				return;
 			}
+			case "ollama-cloud": {
+				const apiKey = await loginOllamaCloud({
+					onAuth(info) {
+						const { url, instructions } = info;
+						console.log(`\nOpen this URL in your browser:\n${url}`);
+						if (instructions) console.log(instructions);
+						console.log();
+					},
+					onPrompt(p) {
+						return promptFn(`${p.message}${p.placeholder ? ` (${p.placeholder})` : ""}:`);
+					},
+				});
+				storage.saveApiKey(provider, apiKey);
+				console.log(`\nAPI key saved to ~/.omp/agent/agent.db`);
+				return;
+			}
+
 			case "minimax-code": {
 				const apiKey = await loginMiniMaxCode({
 					onAuth(info) {
@@ -347,6 +365,7 @@ Providers:
   minimax-code-cn   MiniMax Coding Plan (China)
   cursor            Cursor (Claude, GPT, etc.)
   zenmux            ZenMux
+  ollama-cloud      Ollama Cloud
 
 Examples:
   bunx @oh-my-pi/pi-ai login              # interactive provider selection

package/src/index.ts

@@ -16,6 +16,7 @@ export * from "./providers/google";
 export * from "./providers/google-gemini-cli";
 export * from "./providers/google-vertex";
 export * from "./providers/kimi";
+export * from "./providers/ollama";
 export type { OpenAICodexResponsesOptions } from "./providers/openai-codex-responses";
 export * from "./providers/openai-completions";
 export * from "./providers/openai-responses";

package/src/model-thinking.ts

@@ -158,7 +158,7 @@ export function applyGeneratedModelPolicies(models: ApiModel<Api>[]): void {
  *
  * When a model's context is exhausted, the agent can promote to a sibling
  * model with a larger context window on the same provider:
- * - `-spark` variants promote to `gpt-5.5`.
+ * - `codex-spark` variants promote to `gpt-5.5`.
  * - `gpt-5.5` (270K input) promotes to `gpt-5.4` (1M input).
  */
 export function linkOpenAIPromotionTargets(models: ApiModel<Api>[]): void {
@@ -472,6 +472,9 @@ function inferFallbackEfforts<TApi extends Api>(model: ApiModel<TApi>): readonly
 	if (model.api === "anthropic-messages") {
 		return DEFAULT_REASONING_EFFORTS_WITH_XHIGH;
 	}
+	if (model.name.includes("deepseek-v4")) {
+		return DEFAULT_REASONING_EFFORTS_WITH_XHIGH;
+	}
 	if (model.api === "bedrock-converse-stream") {
 		return DEFAULT_REASONING_EFFORTS;
 	}