feat(ai): add Neuralwatt provider

.agents/skills/senpi-qa/references/env-vars.md

@@ -30,7 +30,7 @@ The first present key wins; `devenv-setup.mjs` seeds `.env.local` from it.
 `MISTRAL_API_KEY`, `MINIMAX_API_KEY`, `MOONSHOT_API_KEY`, `KIMI_API_KEY`,
 `OPENCODE_API_KEY`, `CLOUDFLARE_API_KEY` (+ `CLOUDFLARE_ACCOUNT_ID` /
 `CLOUDFLARE_GATEWAY_ID`), `XIAOMI_API_KEY` (+ regional token-plan keys),
-`ALIBABA_TOKEN_PLAN_API_KEY`,
+`ALIBABA_TOKEN_PLAN_API_KEY`, `NEURALWATT_API_KEY`,
 `HF_TOKEN`, and the AWS Bedrock / Google Vertex variable sets.
 
 OAuth-only subscription providers have no env key: `openai-codex` and

.agents/skills/senpi-qa/scripts/lib/mock-loop-support.mjs

@@ -53,6 +53,7 @@ export const PROVIDER_ENV_KEYS = [
 	"FIREWORKS_API_KEY", "TOGETHER_API_KEY", "OPENROUTER_API_KEY", "AI_GATEWAY_API_KEY", "ZAI_API_KEY",
 	"ZAI_CODING_CN_API_KEY", "MISTRAL_API_KEY", "MINIMAX_API_KEY", "MINIMAX_CN_API_KEY", "MOONSHOT_API_KEY",
 	"MOONSHOTAI_API_KEY", "KIMI_API_KEY", "OPENCODE_API_KEY", "CLOUDFLARE_API_KEY", "HF_TOKEN",
+	"NEURALWATT_API_KEY",
 ];
 
 export function hermeticEnv(boxEnv) {

.devcontainer/devcontainer.json

@@ -31,6 +31,9 @@
 		},
 		"ALIBABA_TOKEN_PLAN_API_KEY": {
 			"description": "(Optional) Alibaba Cloud Model Studio Token Plan API key (ap-southeast-1)."
+		},
+		"NEURALWATT_API_KEY": {
+			"description": "(Optional) Neuralwatt API key (Neuralwatt's OpenAI-compatible inference endpoint)."
 		}
 	},
 	"customizations": {

packages/ai/README.md

@@ -58,6 +58,7 @@ Unified LLM API with provider collections, automatic auth resolution, token and
 
 - **OpenAI**
 - **Ant Ling**
+- **Neuralwatt**
 - **Azure OpenAI (Responses)**
 - **OpenAI Codex** (ChatGPT Plus/Pro subscription, requires OAuth, see below)
 - **DeepSeek**
@@ -419,6 +420,7 @@ Built-in providers resolve these env vars (Node.js; in browsers pass `apiKey` ex
 | OpenAI | `OPENAI_API_KEY` |
 | Ollama Cloud | `OLLAMA_API_KEY` |
 | Ant Ling | `ANT_LING_API_KEY` |
+| Neuralwatt | `NEURALWATT_API_KEY` |
 | Azure OpenAI | `AZURE_OPENAI_API_KEY` + `AZURE_OPENAI_BASE_URL` (e.g. `https://{resource}.ai.azure.com`) or `AZURE_OPENAI_RESOURCE_NAME`. Supports `*.openai.azure.com`, `*.cognitiveservices.azure.com` and `*.ai.azure.com`; root endpoints auto-normalize to `/openai/v1`. Optional: `AZURE_OPENAI_API_VERSION` (default `v1`), `AZURE_OPENAI_DEPLOYMENT_NAME_MAP`. |
 | Anthropic | `ANTHROPIC_API_KEY` or `ANTHROPIC_OAUTH_TOKEN` |
 | DeepSeek | `DEEPSEEK_API_KEY` |

packages/ai/scripts/generate-models.ts

@@ -3038,6 +3038,302 @@ async function generateModels() {
 		}
 	}
 
+	const neuralwattCompat: OpenAICompletionsCompat = {
+		supportsStore: false,
+		supportsDeveloperRole: false,
+		supportsReasoningEffort: false,
+		maxTokensField: "max_tokens",
+	};
+	const neuralwattModels: Model<"openai-completions">[] = [
+		{
+			id: "qwen3.6-35b-fast",
+			name: "Qwen3.6 35B Fast",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text", "image"],
+			cost: { input: 0.29, output: 1.15, cacheRead: 0.0725, cacheWrite: 0 },
+			contextWindow: 131056,
+			maxTokens: 131056,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "kimi-k3",
+			name: "Kimi K3",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 3, output: 15, cacheRead: 0.3, cacheWrite: 0 },
+			contextWindow: 1048560,
+			maxTokens: 1048560,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "deepseek-v4-flash",
+			name: "DeepSeek V4 Flash",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 0.104, output: 0.207, cacheRead: 0.026, cacheWrite: 0 },
+			contextWindow: 1048560,
+			maxTokens: 65536,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "kimi-k3-fast",
+			name: "Kimi K3 Fast",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text", "image"],
+			cost: { input: 3, output: 15, cacheRead: 0.3, cacheWrite: 0 },
+			contextWindow: 1048560,
+			maxTokens: 1048560,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "glm-5.2-fast",
+			name: "GLM 5.2 Fast",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 1.45, output: 4.5, cacheRead: 0.3625, cacheWrite: 0 },
+			contextWindow: 1048560,
+			maxTokens: 1048560,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "glm-5.2-flex",
+			name: "GLM 5.2 Flex",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 0.725, output: 2.25, cacheRead: 0.18125, cacheWrite: 0 },
+			contextWindow: 1048560,
+			maxTokens: 1048560,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "glm-5.2-short",
+			name: "GLM 5.2 Short",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 1.45, output: 4.5, cacheRead: 0.3625, cacheWrite: 0 },
+			contextWindow: 199984,
+			maxTokens: 199984,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "gemma-4-31b",
+			name: "Gemma 4 31B",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 0.144, output: 0.42, cacheRead: 0.036, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 16384,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "kimi-k2.5-fast",
+			name: "Kimi K2.5 Fast",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text", "image"],
+			cost: { input: 0.52, output: 2.59, cacheRead: 0.13, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 262128,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "kimi-k2.6-flex",
+			name: "Kimi K2.6 Flex",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 0.345, output: 1.61, cacheRead: 0.08625, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 262128,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "qwen3.5-397b-fast",
+			name: "Qwen3.5 397B Fast",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 0.69, output: 4.14, cacheRead: 0.1725, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 262128,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "glm-5.2-short-fast-flex",
+			name: "GLM 5.2 Short Fast Flex",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 0.725, output: 2.25, cacheRead: 0.18125, cacheWrite: 0 },
+			contextWindow: 199984,
+			maxTokens: 199984,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "glm-5.2",
+			name: "GLM 5.2",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 1.45, output: 4.5, cacheRead: 0.3625, cacheWrite: 0 },
+			contextWindow: 1048560,
+			maxTokens: 1048560,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "kimi-k2.6-fast",
+			name: "Kimi K2.6 Fast",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text", "image"],
+			cost: { input: 0.69, output: 3.22, cacheRead: 0.1725, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 262128,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "glm-5.2-short-fast",
+			name: "GLM 5.2 Short Fast",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 1.45, output: 4.5, cacheRead: 0.3625, cacheWrite: 0 },
+			contextWindow: 199984,
+			maxTokens: 199984,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "kimi-k2.7-code-flex",
+			name: "Kimi K2.7 Code Flex",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 0.475, output: 2, cacheRead: 0.11875, cacheWrite: 0 },
+			contextWindow: 262144,
+			maxTokens: 262144,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "glm-5.2-short-flex",
+			name: "GLM 5.2 Short Flex",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 0.725, output: 2.25, cacheRead: 0.18125, cacheWrite: 0 },
+			contextWindow: 199984,
+			maxTokens: 199984,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "moonshotai/Kimi-K2.5",
+			name: "Kimi K2.5",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 0.52, output: 2.59, cacheRead: 0.13, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 262128,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "moonshotai/Kimi-K2.7-Code",
+			name: "Kimi K2.7 Code",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 0.95, output: 4, cacheRead: 0.2375, cacheWrite: 0 },
+			contextWindow: 262144,
+			maxTokens: 262144,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "moonshotai/Kimi-K2.6",
+			name: "Kimi K2.6",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 0.69, output: 3.22, cacheRead: 0.1725, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 262128,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "Qwen/Qwen3.5-397B-A17B-FP8",
+			name: "Qwen3.5 397B A17B FP8",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 0.69, output: 4.14, cacheRead: 0.1725, cacheWrite: 0 },
+			contextWindow: 262128,
+			maxTokens: 262128,
+			compat: neuralwattCompat,
+		},
+		{
+			id: "Qwen/Qwen3.6-35B-A3B",
+			name: "Qwen3.6 35B A3B",
+			api: "openai-completions",
+			baseUrl: "https://api.neuralwatt.com/v1",
+			provider: "neuralwatt",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: { input: 0.29, output: 1.15, cacheRead: 0.0725, cacheWrite: 0 },
+			contextWindow: 131056,
+			maxTokens: 131056,
+			compat: neuralwattCompat,
+		},
+	];
+	allModels.push(...neuralwattModels);
+
 	const minimaxDirectSupportedIds = new Set(["MiniMax-M2.7", "MiniMax-M2.7-highspeed", "MiniMax-M3"]);
 
 	for (let i = allModels.length - 1; i >= 0; i--) {

packages/ai/src/changes.md

@@ -184,6 +184,49 @@ The divergence lives in core wiring, package identity, or build plumbing that ex
 ### Expected merge conflict zones
 
 - LOW: Cursor `resource_exhausted` handling in `packages/ai/src/utils/overflow.ts`.
+## 2026-08-21 - Add Neuralwatt provider
+
+### What changed
+
+- `packages/ai/scripts/generate-models.ts`: 22-model `neuralwatt` catalog block (`openai-completions`
+  API, ant-ling entry pattern).
+- `packages/ai/src/env-api-keys.ts`: maps `NEURALWATT_API_KEY`.
+- `packages/ai/src/types.ts`: `neuralwatt` joins the `KnownProvider` union.
+- `packages/ai/src/providers/all.ts`: registers the Neuralwatt factory in the `builtinProviders()` list.
+- `packages/ai/src/providers/neuralwatt.ts` and `packages/ai/src/providers/neuralwatt.models.ts` (new):
+  provider factory and catalog module.
+- `packages/ai/README.md`: Neuralwatt bullet in the provider list and a `NEURALWATT_API_KEY` row in the
+  env-var matrix.
+- Test files: provider unit/request tests and env-key coverage under `packages/ai/test/`,
+  model-resolver coverage under `packages/coding-agent/test/`.
+- Coding-agent mirror files: the `Record<KnownProvider, ...>` maps and display-name entries in
+  `packages/coding-agent/src/cli/args.ts`, `src/core/model-config.ts`, `src/core/model-config-schema.ts`,
+  `src/core/model-resolver.ts`, `src/core/provider-display-names.ts`, plus
+  `packages/coding-agent/docs/providers.md`.
+- Infra mirror files: `.devcontainer/devcontainer.json`, the `test.sh` / `pi-test.sh` env-unset lists,
+  and the senpi-qa skill env-vars reference / mock-loop support.
+- Generated artifacts: `packages/ai/src/models.generated.ts` and `packages/ai/src/providers/data/`
+  (including `neuralwatt.json`) — resolve by regenerating.
+
+### Why
+
+- Neuralwatt is a public OpenAI-compatible Korean inference API (docs:
+  https://portal.neuralwatt.com/docs, free tier) that omo-native/senpi users currently cannot reach
+  without per-user `models.json` overlays; adding it as a built-in provider lets users configure only
+  an API key.
+
+### Why an extension could not handle it
+
+- Built-in provider catalogs and env-key detection live in `packages/ai` core; extensions cannot
+  register a builtin catalog provider or contribute to the `KnownProvider` union /
+  `builtinProviders()` map / env-api-keys map.
+
+### Expected merge conflict zones
+
+- `packages/ai/src/env-api-keys.ts` env-key map.
+- `packages/ai/src/providers/all.ts` `builtinProviders` list.
+- `packages/ai/src/types.ts` `KnownProvider` union.
+- `packages/ai/scripts/generate-models.ts` ant-ling-adjacent block.
 
 ## Unreleased