Skip to content

Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.

Add Tinfoil private inference provider (#875)

Zach880f4f6c10ef
diff --git a/.env.example b/.env.example
index 8bc9e3b7a..7a3712042 100644
--- a/.env.example
+++ b/.env.example
@@ -32,6 +32,10 @@ GEMINI_API_KEY=your_gemini_api_key_here
# Get your API key from: https://console.groq.com/keys
GROQ_API_KEY=your_groq_api_key_here
+# Tinfoil API Configuration (optional - for private cloud LLM inference)
+# Get your API key from: https://tinfoil.sh/inference
+TINFOIL_API_KEY=your_tinfoil_api_key_here
+
# Mistral API Configuration (optional - for Voxtral transcription)
# Get your API key from: https://console.mistral.ai/api-keys
MISTRAL_API_KEY=your_mistral_api_key_here
diff --git a/README.md b/README.md
index d4fd8cf5a..3daf91ee2 100644
--- a/README.md
+++ b/README.md
@@ -41,7 +41,7 @@ OpenWhispr turns your voice into text, notes, and actions from your desktop. Pre
## Features
- **Voice dictation** — global hotkey to dictate into any app with automatic pasting
-- **AI agent** — talk to GPT-5, Claude, Gemini, Groq, or local models with a named voice assistant
+- **AI agent** — talk to GPT-5, Claude, Gemini, Groq, Tinfoil, or local models with a named voice assistant
- **Voice agent hotkey** — dedicated hotkey that sends your dictation straight to your AI agent as a command, no wake word needed and no cleanup pass
- **Meeting transcription** — auto-detect Zoom, Teams, and FaceTime calls with live speaker diarization, voice fingerprinting, and Google Calendar integration
- **Local speaker diarization** — on-device speaker labelling with voice fingerprint recognition across meetings, no cloud required
diff --git a/docs/network-allowlist.md b/docs/network-allowlist.md
index 5395cd745..2d8166cd8 100644
--- a/docs/network-allowlist.md
+++ b/docs/network-allowlist.md
@@ -54,13 +54,14 @@ Contacted only if the user connects Google Calendar in settings.
Required only when a user configures their own API key for the corresponding
provider. Skip any provider not in use.
-| Host | Protocol | Port | Used when |
-| ----------------------------------- | -------- | ---- | ------------------------------------------------------- |
-| `api.openai.com` | HTTPS | 443 | OpenAI API key configured (transcription or reasoning). |
-| `api.anthropic.com` | HTTPS | 443 | Anthropic API key configured. |
-| `generativelanguage.googleapis.com` | HTTPS | 443 | Gemini API key configured. |
-| `api.groq.com` | HTTPS | 443 | Groq API key configured. |
-| `api.mistral.ai` | HTTPS | 443 | Mistral API key configured. |
+| Host | Protocol | Port | Used when |
+| ----------------------------------- | ---------- | ---- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `api.openai.com` | HTTPS | 443 | OpenAI API key configured (transcription or reasoning). |
+| `api.anthropic.com` | HTTPS | 443 | Anthropic API key configured. |
+| `generativelanguage.googleapis.com` | HTTPS | 443 | Gemini API key configured. |
+| `api.groq.com` | HTTPS | 443 | Groq API key configured. |
+| `atc.tinfoil.sh`, `*.tinfoil.sh` | WSS, HTTPS | 443 | Tinfoil API key configured. `atc.tinfoil.sh` serves the enclave attestation bundle (verified locally against an embedded sigstore root). Inference and realtime transcription connect to an enclave host assigned dynamically at runtime (e.g. `inference.tinfoil.sh`, `router.infN.tinfoil.sh`), so allowlist `*.tinfoil.sh` rather than pinning hosts. |
+| `api.mistral.ai` | HTTPS | 443 | Mistral API key configured. |
## Notes
diff --git a/src/components/ReasoningModelSelector.tsx b/src/components/ReasoningModelSelector.tsx
index e0150d970..97601221a 100644
--- a/src/components/ReasoningModelSelector.tsx
+++ b/src/components/ReasoningModelSelector.tsx
@@ -33,7 +33,7 @@ type CloudModelOption = {
invertInDark?: boolean;
};
-const CLOUD_PROVIDER_IDS = ["openai", "anthropic", "gemini", "groq", "custom"];
+const CLOUD_PROVIDER_IDS = ["openai", "anthropic", "gemini", "groq", "tinfoil", "custom"];
interface ReasoningModelSelectorProps {
reasoningModel: string;
@@ -320,6 +320,8 @@ export default function ReasoningModelSelector({
const setGeminiApiKey = useSettingsStore((s) => s.setGeminiApiKey);
const groqApiKey = useSettingsStore((s) => s.groqApiKey);
const setGroqApiKey = useSettingsStore((s) => s.setGroqApiKey);
+ const tinfoilApiKey = useSettingsStore((s) => s.tinfoilApiKey);
+ const setTinfoilApiKey = useSettingsStore((s) => s.setTinfoilApiKey);
const [selectedMode, setSelectedMode] = useState<"cloud" | "local">(mode || "cloud");
const [selectedCloudProvider, setSelectedCloudProvider] = useState("openai");
const [selectedLocalProvider, setSelectedLocalProvider] = useState("qwen");
@@ -624,6 +626,31 @@ export default function ReasoningModelSelector({
</div>
)}
+ {selectedCloudProvider === "tinfoil" && (
+ <div className="space-y-2">
+ <div className="flex items-baseline justify-between">
+ <h4 className="font-medium text-foreground">{t("common.apiKey")}</h4>
+ <a
+ href="https://tinfoil.sh/inference?utm_source=referral&utm_campaign=openwhispr"
+ target="_blank"
+ rel="noopener noreferrer"
+ onClick={createExternalLinkHandler(
+ "https://tinfoil.sh/inference?utm_source=referral&utm_campaign=openwhispr"
+ )}
+ className="text-xs text-link underline decoration-link/30 hover:decoration-link/60 cursor-pointer transition-colors"
+ >
+ {t("reasoning.getApiKey")}
+ </a>
+ </div>
+ <ApiKeyInput
+ apiKey={tinfoilApiKey}
+ setApiKey={setTinfoilApiKey}
+ label=""
+ helpText=""
+ />
+ </div>
+ )}
+
<div className="pt-3 space-y-2">
<h4 className="text-sm font-medium text-foreground">
{t("reasoning.selectModel")}
diff --git a/src/components/chat/useChatStreaming.ts b/src/components/chat/useChatStreaming.ts
index 0b3420103..9484114c7 100644
--- a/src/components/chat/useChatStreaming.ts
+++ b/src/components/chat/useChatStreaming.ts
@@ -99,9 +99,14 @@ export function useChatStreaming({
const isLanAgent = chatAgentMode === "self-hosted" && !!settings.chatAgentRemoteUrl;
const isCustomAgent =
chatAgentMode === "providers" && settings.chatAgentProvider === "custom";
- const isLocalProvider = !["openai", "groq", "custom", "anthropic", "gemini"].includes(
- settings.chatAgentProvider
- );
+ const isLocalProvider = ![
+ "openai",
+ "groq",
+ "custom",
+ "anthropic",
+ "gemini",
+ "tinfoil",
+ ].includes(settings.chatAgentProvider);
const localModelCanUseTool =
isLocalProvider && estimateModelSizeB(settings.chatAgentModel) >= LOCAL_TOOL_MIN_PARAMS_B;
const supportsTools = isCloudAgent || !isLocalProvider || localModelCanUseTool;
diff --git a/src/components/ui/PromptStudio.tsx b/src/components/ui/PromptStudio.tsx
index eaa95c490..e45d2347e 100644
--- a/src/components/ui/PromptStudio.tsx
+++ b/src/components/ui/PromptStudio.tsx
@@ -38,6 +38,7 @@ const PROVIDER_CONFIG: Record<string, ProviderConfig> = {
anthropic: { label: "Anthropic", apiKeyStorageKey: "anthropicApiKey" },
gemini: { label: "Gemini", apiKeyStorageKey: "geminiApiKey" },
groq: { label: "Groq", apiKeyStorageKey: "groqApiKey" },
+ tinfoil: { label: "Tinfoil", apiKeyStorageKey: "tinfoilApiKey" },
openwhispr: { label: "OpenWhispr Cloud" },
custom: {
label: "Custom endpoint",
diff --git a/src/locales/de/translation.json b/src/locales/de/translation.json
index 62941e205..246473bfd 100644
--- a/src/locales/de/translation.json
+++ b/src/locales/de/translation.json
@@ -1894,7 +1894,13 @@
"anthropic_claude_opus_4_8": "Leistungsstärkstes Claude-Modell",
"anthropic_claude_opus_4_7": "Leistungsstarkes Opus-Modell",
"anthropic_claude_opus_4_6": "Vorherige Opus-Generation, 1M Kontext",
- "anthropic_claude_opus_4_5": "Früheres Opus-Modell"
+ "anthropic_claude_opus_4_5": "Früheres Opus-Modell",
+ "tinfoil_deepseek_v4_pro": "Privates Long-Context-Reasoning in sicheren Enklaven, 800K Kontext",
+ "tinfoil_glm_5_2": "Privates agentisches Engineering-Modell in sicheren Enklaven, 384K Kontext",
+ "tinfoil_kimi_k2_6": "Privates Modell für langfristiges Coding und Multimodalität, 256K Kontext",
+ "tinfoil_gemma4_31b": "Privates mehrsprachiges Reasoning-Modell mit Tool-Aufrufen, 256K Kontext",
+ "tinfoil_gpt_oss_120b": "Privates Open-Weight-Reasoning-Modell mit Tool-Nutzung, 131K Kontext",
+ "tinfoil_llama3_3_70b": "Privates Konversationsmodell mit Tool-Aufrufen, 128K Kontext"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "Schnell und günstig. Ideal zum Bereinigen von Text.",
diff --git a/src/locales/en/translation.json b/src/locales/en/translation.json
index 5ba722519..0dc22a14e 100644
--- a/src/locales/en/translation.json
+++ b/src/locales/en/translation.json
@@ -1986,7 +1986,13 @@
"anthropic_claude_opus_4_8": "Most capable Claude model",
"anthropic_claude_opus_4_7": "Powerful Opus model",
"anthropic_claude_opus_4_6": "Previous Opus generation, 1M context",
- "anthropic_claude_opus_4_5": "Earlier Opus model"
+ "anthropic_claude_opus_4_5": "Earlier Opus model",
+ "tinfoil_deepseek_v4_pro": "Private long-context reasoning in secure enclaves, 800K context",
+ "tinfoil_glm_5_2": "Private agentic engineering model in secure enclaves, 384K context",
+ "tinfoil_kimi_k2_6": "Private long-horizon coding and multimodal model, 256K context",
+ "tinfoil_gemma4_31b": "Private multilingual reasoning model with tool calling, 256K context",
+ "tinfoil_gpt_oss_120b": "Private open-weight reasoning model with tool use, 131K context",
+ "tinfoil_llama3_3_70b": "Private conversational model with tool calling, 128K context"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "Fast and cheap. Great default for text cleanup.",
diff --git a/src/locales/es/translation.json b/src/locales/es/translation.json
index 50840e640..3e24210d4 100644
--- a/src/locales/es/translation.json
+++ b/src/locales/es/translation.json
@@ -1942,7 +1942,13 @@
"anthropic_claude_opus_4_8": "Modelo Claude más capaz",
"anthropic_claude_opus_4_7": "Modelo Opus potente",
"anthropic_claude_opus_4_6": "Generación anterior de Opus, contexto de 1M",
- "anthropic_claude_opus_4_5": "Modelo Opus anterior"
+ "anthropic_claude_opus_4_5": "Modelo Opus anterior",
+ "tinfoil_deepseek_v4_pro": "Razonamiento privado de contexto largo en enclaves seguros, contexto de 800K",
+ "tinfoil_glm_5_2": "Modelo privado de ingeniería agéntica en enclaves seguros, contexto de 384K",
+ "tinfoil_kimi_k2_6": "Modelo privado de codificación a largo plazo y multimodal, contexto de 256K",
+ "tinfoil_gemma4_31b": "Modelo privado de razonamiento multilingüe con llamada a herramientas, contexto de 256K",
+ "tinfoil_gpt_oss_120b": "Modelo privado de razonamiento de pesos abiertos con uso de herramientas, contexto de 131K",
+ "tinfoil_llama3_3_70b": "Modelo privado conversacional con llamada a herramientas, contexto de 128K"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "Rápido y económico. Ideal para limpiar texto.",
diff --git a/src/locales/fr/translation.json b/src/locales/fr/translation.json
index 1ca1adef6..05e0a1cac 100644
--- a/src/locales/fr/translation.json
+++ b/src/locales/fr/translation.json
@@ -1942,7 +1942,13 @@
"anthropic_claude_opus_4_8": "Modèle Claude le plus performant",
"anthropic_claude_opus_4_7": "Modèle Opus puissant",
"anthropic_claude_opus_4_6": "Génération Opus précédente, contexte 1M",
- "anthropic_claude_opus_4_5": "Modèle Opus antérieur"
+ "anthropic_claude_opus_4_5": "Modèle Opus antérieur",
+ "tinfoil_deepseek_v4_pro": "Raisonnement privé à long contexte dans des enclaves sécurisées, contexte 800K",
+ "tinfoil_glm_5_2": "Modèle privé d'ingénierie agentique dans des enclaves sécurisées, contexte 384K",
+ "tinfoil_kimi_k2_6": "Modèle privé de codage longue durée et multimodal, contexte 256K",
+ "tinfoil_gemma4_31b": "Modèle privé de raisonnement multilingue avec appel d'outils, contexte 256K",
+ "tinfoil_gpt_oss_120b": "Modèle privé de raisonnement à poids ouverts avec usage d'outils, contexte 131K",
+ "tinfoil_llama3_3_70b": "Modèle privé conversationnel avec appel d'outils, contexte 128K"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "Rapide et économique. Parfait pour nettoyer le texte.",
diff --git a/src/locales/it/translation.json b/src/locales/it/translation.json
index 118e1a188..d423fe6e7 100644
--- a/src/locales/it/translation.json
+++ b/src/locales/it/translation.json
@@ -1894,7 +1894,13 @@
"anthropic_claude_opus_4_8": "Modello Claude più capace",
"anthropic_claude_opus_4_7": "Modello Opus potente",
"anthropic_claude_opus_4_6": "Generazione Opus precedente, contesto da 1M",
- "anthropic_claude_opus_4_5": "Modello Opus meno recente"
+ "anthropic_claude_opus_4_5": "Modello Opus meno recente",
+ "tinfoil_deepseek_v4_pro": "Ragionamento privato a contesto lungo in enclave sicure, contesto 800K",
+ "tinfoil_glm_5_2": "Modello privato di ingegneria agentica in enclave sicure, contesto 384K",
+ "tinfoil_kimi_k2_6": "Modello privato di coding a lungo termine e multimodale, contesto 256K",
+ "tinfoil_gemma4_31b": "Modello privato di ragionamento multilingue con chiamata di strumenti, contesto 256K",
+ "tinfoil_gpt_oss_120b": "Modello privato di ragionamento open-weight con uso di strumenti, contesto 131K",
+ "tinfoil_llama3_3_70b": "Modello privato conversazionale con chiamata di strumenti, contesto 128K"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "Veloce ed economico. Ideale per ripulire il testo.",
diff --git a/src/locales/ja/translation.json b/src/locales/ja/translation.json
index e70e63ddc..b3ddb822f 100644
--- a/src/locales/ja/translation.json
+++ b/src/locales/ja/translation.json
@@ -1894,7 +1894,13 @@
"anthropic_claude_opus_4_8": "最も高性能な Claude モデル",
"anthropic_claude_opus_4_7": "高性能な Opus モデル",
"anthropic_claude_opus_4_6": "前世代の Opus、100 万トークンコンテキスト",
- "anthropic_claude_opus_4_5": "旧世代の Opus モデル"
+ "anthropic_claude_opus_4_5": "旧世代の Opus モデル",
+ "tinfoil_deepseek_v4_pro": "セキュアエンクレーブ内のプライベート長文脈推論、800K コンテキスト",
+ "tinfoil_glm_5_2": "セキュアエンクレーブ内のプライベート・エージェント型エンジニアリングモデル、384K コンテキスト",
+ "tinfoil_kimi_k2_6": "プライベートな長期コーディング・マルチモーダルモデル、256K コンテキスト",
+ "tinfoil_gemma4_31b": "ツール呼び出し対応のプライベート多言語推論モデル、256K コンテキスト",
+ "tinfoil_gpt_oss_120b": "ツール使用対応のプライベート・オープンウェイト推論モデル、131K コンテキスト",
+ "tinfoil_llama3_3_70b": "ツール呼び出し対応のプライベート会話モデル、128K コンテキスト"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "高速で低コスト。テキスト整形に最適です。",
diff --git a/src/locales/pt/translation.json b/src/locales/pt/translation.json
index 9e3554e20..248c967c6 100644
--- a/src/locales/pt/translation.json
+++ b/src/locales/pt/translation.json
@@ -1873,7 +1873,13 @@
"anthropic_claude_opus_4_8": "Modelo Claude mais capaz",
"anthropic_claude_opus_4_7": "Modelo Opus potente",
"anthropic_claude_opus_4_6": "Geração anterior do Opus, contexto de 1M",
- "anthropic_claude_opus_4_5": "Modelo Opus mais antigo"
+ "anthropic_claude_opus_4_5": "Modelo Opus mais antigo",
+ "tinfoil_deepseek_v4_pro": "Raciocínio privado de contexto longo em enclaves seguros, contexto de 800K",
+ "tinfoil_glm_5_2": "Modelo privado de engenharia agêntica em enclaves seguros, contexto de 384K",
+ "tinfoil_kimi_k2_6": "Modelo privado de codificação de longo prazo e multimodal, contexto de 256K",
+ "tinfoil_gemma4_31b": "Modelo privado de raciocínio multilíngue com chamada de ferramentas, contexto de 256K",
+ "tinfoil_gpt_oss_120b": "Modelo privado de raciocínio de pesos abertos com uso de ferramentas, contexto de 131K",
+ "tinfoil_llama3_3_70b": "Modelo privado conversacional com chamada de ferramentas, contexto de 128K"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "Rápido e barato. Ótimo para limpar texto.",
diff --git a/src/locales/ru/translation.json b/src/locales/ru/translation.json
index 8d71af0b2..e950197d2 100644
--- a/src/locales/ru/translation.json
+++ b/src/locales/ru/translation.json
@@ -1894,7 +1894,13 @@
"anthropic_claude_opus_4_8": "Самая мощная модель Claude",
"anthropic_claude_opus_4_7": "Мощная модель Opus",
"anthropic_claude_opus_4_6": "Предыдущее поколение Opus, контекст 1M",
- "anthropic_claude_opus_4_5": "Более ранняя модель Opus"
+ "anthropic_claude_opus_4_5": "Более ранняя модель Opus",
+ "tinfoil_deepseek_v4_pro": "Приватное рассуждение с длинным контекстом в защищённых анклавах, контекст 800K",
+ "tinfoil_glm_5_2": "Приватная агентная инженерная модель в защищённых анклавах, контекст 384K",
+ "tinfoil_kimi_k2_6": "Приватная модель для длительного кодинга и мультимодальности, контекст 256K",
+ "tinfoil_gemma4_31b": "Приватная многоязычная модель рассуждений с вызовом инструментов, контекст 256K",
+ "tinfoil_gpt_oss_120b": "Приватная модель рассуждений с открытыми весами и использованием инструментов, контекст 131K",
+ "tinfoil_llama3_3_70b": "Приватная разговорная модель с вызовом инструментов, контекст 128K"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "Быстрая и дешёвая. Отлично подходит для очистки текста.",
diff --git a/src/locales/zh-CN/translation.json b/src/locales/zh-CN/translation.json
index ee4bde64d..c9c638624 100644
--- a/src/locales/zh-CN/translation.json
+++ b/src/locales/zh-CN/translation.json
@@ -1889,7 +1889,13 @@
"anthropic_claude_opus_4_8": "最强大的 Claude 模型",
"anthropic_claude_opus_4_7": "强大的 Opus 模型",
"anthropic_claude_opus_4_6": "上一代 Opus,1M 上下文",
- "anthropic_claude_opus_4_5": "更早的 Opus 模型"
+ "anthropic_claude_opus_4_5": "更早的 Opus 模型",
+ "tinfoil_deepseek_v4_pro": "安全飞地中的私密长上下文推理,800K 上下文",
+ "tinfoil_glm_5_2": "安全飞地中的私密智能体工程模型,384K 上下文",
+ "tinfoil_kimi_k2_6": "私密长程编码与多模态模型,256K 上下文",
+ "tinfoil_gemma4_31b": "支持工具调用的私密多语言推理模型,256K 上下文",
+ "tinfoil_gpt_oss_120b": "支持工具使用的私密开源权重推理模型,131K 上下文",
+ "tinfoil_llama3_3_70b": "支持工具调用的私密对话模型,128K 上下文"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "快速且低成本。适合文本清理。",
diff --git a/src/locales/zh-TW/translation.json b/src/locales/zh-TW/translation.json
index ceb3b2e33..83e56bbbc 100644
--- a/src/locales/zh-TW/translation.json
+++ b/src/locales/zh-TW/translation.json
@@ -1889,7 +1889,13 @@
"anthropic_claude_opus_4_8": "最強大的 Claude 模型",
"anthropic_claude_opus_4_7": "強大的 Opus 模型",
"anthropic_claude_opus_4_6": "上一代 Opus 模型,1M 上下文",
- "anthropic_claude_opus_4_5": "較早的 Opus 模型"
+ "anthropic_claude_opus_4_5": "較早的 Opus 模型",
+ "tinfoil_deepseek_v4_pro": "安全飛地中的私密長上下文推理,800K 上下文",
+ "tinfoil_glm_5_2": "安全飛地中的私密智能體工程模型,384K 上下文",
+ "tinfoil_kimi_k2_6": "私密長程編碼與多模態模型,256K 上下文",
+ "tinfoil_gemma4_31b": "支援工具呼叫的私密多語言推理模型,256K 上下文",
+ "tinfoil_gpt_oss_120b": "支援工具使用的私密開放權重推理模型,131K 上下文",
+ "tinfoil_llama3_3_70b": "支援工具呼叫的私密對話模型,128K 上下文"
},
"enterprise": {
"bedrock_claude_haiku_4_5": "快速且低成本。適合文字清理。",
diff --git a/src/models/modelRegistryData.json b/src/models/modelRegistryData.json
index 7f453802c..b12b10f43 100644
--- a/src/models/modelRegistryData.json
+++ b/src/models/modelRegistryData.json
@@ -451,6 +451,65 @@
"descriptionKey": "models.descriptions.cloud.groq_kimi_k2_0905"
}
]
+ },
+ {
+ "id": "tinfoil",
+ "name": "Tinfoil",
+ "models": [
+ {
+ "id": "deepseek-v4-pro",
+ "name": "DeepSeek V4 Pro",
+ "description": "Private long-context reasoning in secure enclaves, 800K context",
+ "tokenParam": "max_tokens",
+ "supportsTemperature": true,
+ "descriptionKey": "models.descriptions.cloud.tinfoil_deepseek_v4_pro",
+ "supportsThinking": true
+ },
+ {
+ "id": "glm-5-2",
+ "name": "GLM-5.2",
+ "description": "Private agentic engineering model in secure enclaves, 384K context",
+ "tokenParam": "max_tokens",
+ "supportsTemperature": true,
+ "descriptionKey": "models.descriptions.cloud.tinfoil_glm_5_2",
+ "supportsThinking": true
+ },
+ {
+ "id": "kimi-k2-6",
+ "name": "Kimi K2.6",
+ "description": "Private long-horizon coding and multimodal model, 256K context",
+ "tokenParam": "max_tokens",
+ "supportsTemperature": true,
+ "descriptionKey": "models.descriptions.cloud.tinfoil_kimi_k2_6",
+ "supportsThinking": true
+ },
+ {
+ "id": "gemma4-31b",
+ "name": "Gemma 4 31B",
+ "description": "Private multilingual reasoning model with tool calling, 256K context",
+ "tokenParam": "max_tokens",
+ "supportsTemperature": true,
+ "descriptionKey": "models.descriptions.cloud.tinfoil_gemma4_31b",
+ "supportsThinking": true
+ },
+ {
+ "id": "gpt-oss-120b",
+ "name": "GPT-OSS 120B",
+ "description": "Private open-weight reasoning model with tool use, 131K context",
+ "tokenParam": "max_tokens",
+ "supportsTemperature": true,
+ "descriptionKey": "models.descriptions.cloud.tinfoil_gpt_oss_120b",
+ "supportsThinking": true
+ },
+ {
+ "id": "llama3-3-70b",
+ "name": "Llama 3.3 70B",
+ "description": "Private conversational model with tool calling, 128K context",
+ "tokenParam": "max_tokens",
+ "supportsTemperature": true,
+ "descriptionKey": "models.descriptions.cloud.tinfoil_llama3_3_70b"
+ }
+ ]
}
],
"enterpriseProviders": [
diff --git a/src/services/ReasoningService.ts b/src/services/ReasoningService.ts
index a9f7df3e2..6a0dce77f 100644
--- a/src/services/ReasoningService.ts
+++ b/src/services/ReasoningService.ts
@@ -15,6 +15,7 @@ import { getAIModel } from "./ai/providers";
import { PROVIDER_REGISTRY, type ProviderContext } from "./ai/inferenceProviders";
import { getConfiguredOpenAIBase } from "./ai/openaiBase";
import { applyThinkingSuppression } from "./ai/thinkingSuppression";
+import { clearTinfoilClientCache } from "./ai/tinfoilClient";
export type AgentStreamChunk =
| { type: "content"; text: string }
@@ -62,7 +63,7 @@ class ReasoningService extends BaseReasoningService {
}
private async getApiKey(
- provider: "openai" | "anthropic" | "gemini" | "groq" | "custom"
+ provider: "openai" | "anthropic" | "gemini" | "groq" | "tinfoil" | "custom"
): Promise<string> {
if (provider === "custom") {
let customKey = "";
@@ -100,6 +101,7 @@ class ReasoningService extends BaseReasoningService {
anthropic: () => window.electronAPI.getAnthropicKey(),
gemini: () => window.electronAPI.getGeminiKey(),
groq: () => window.electronAPI.getGroqKey(),
+ tinfoil: () => window.electronAPI.getTinfoilKey?.(),
};
apiKey = (await keyGetters[provider]()) ?? undefined;
@@ -335,7 +337,7 @@ class ReasoningService extends BaseReasoningService {
provider: string,
config: ReasoningConfig & { systemPrompt: string }
): AsyncGenerator<string, void, unknown> {
- const cloudProviders = ["openai", "groq", "gemini", "anthropic", "custom"];
+ const cloudProviders = ["openai", "groq", "gemini", "anthropic", "tinfoil", "custom"];
const isLocalProvider = !cloudProviders.includes(provider);
const settings = getSettings();
@@ -356,7 +358,13 @@ class ReasoningService extends BaseReasoningService {
}
endpoint = `http://127.0.0.1:${serverResult.port}/v1/chat/completions`;
} else {
- const providerKey = provider as "openai" | "groq" | "gemini" | "anthropic" | "custom";
+ const providerKey = provider as
+ | "openai"
+ | "groq"
+ | "gemini"
+ | "anthropic"
+ | "tinfoil"
+ | "custom";
const overrideKey = providerKey === "custom" ? config.customApiKey?.trim() : "";
apiKey = overrideKey || (await this.getApiKey(providerKey));
@@ -367,6 +375,8 @@ class ReasoningService extends BaseReasoningService {
case "gemini":
endpoint = buildApiUrl(API_ENDPOINTS.GEMINI, "/openai/chat/completions");
break;
+ case "tinfoil":
+ throw new Error("Tinfoil streaming must use the verified SDK transport");
case "openai":
case "custom":
endpoint = buildApiUrl(
@@ -538,7 +548,7 @@ class ReasoningService extends BaseReasoningService {
);
}
- const cloudProviders = ["openai", "groq", "gemini", "anthropic", "custom"];
+ const cloudProviders = ["openai", "groq", "gemini", "anthropic", "tinfoil", "custom"];
const isLocalProvider = !cloudProviders.includes(provider);
const settings = getSettings();
@@ -567,7 +577,13 @@ class ReasoningService extends BaseReasoningService {
}
baseURL = `http://127.0.0.1:${serverResult.port}/v1`;
} else {
- const providerKey = provider as "openai" | "groq" | "gemini" | "anthropic" | "custom";
+ const providerKey = provider as
+ | "openai"
+ | "groq"
+ | "gemini"
+ | "anthropic"
+ | "tinfoil"
+ | "custom";
const overrideKey = providerKey === "custom" ? config.customApiKey?.trim() : "";
apiKey = overrideKey || (await this.getApiKey(providerKey));
baseURL =
@@ -576,7 +592,7 @@ class ReasoningService extends BaseReasoningService {
const apiConfig = getOpenAiApiConfig(model);
const aiProvider = isLocalProvider || isLanCleanup ? "local" : provider;
- const aiModel = getAIModel(aiProvider, model, apiKey, baseURL);
+ const aiModel = await getAIModel(aiProvider, model, apiKey, baseURL);
const modelDef = getCloudModel(model);
const userSuppressesThinking = config.disableThinking === true && !!modelDef?.supportsThinking;
@@ -858,6 +874,7 @@ class ReasoningService extends BaseReasoningService {
const anthropicKey = await window.electronAPI?.getAnthropicKey?.();
const geminiKey = await window.electronAPI?.getGeminiKey?.();
const groqKey = await window.electronAPI?.getGroqKey?.();
+ const tinfoilKey = await window.electronAPI?.getTinfoilKey?.();
const localAvailable = await window.electronAPI?.checkLocalReasoningAvailable?.();
logger.logReasoning("API_KEY_CHECK", {
@@ -865,10 +882,11 @@ class ReasoningService extends BaseReasoningService {
hasAnthropic: !!anthropicKey,
hasGemini: !!geminiKey,
hasGroq: !!groqKey,
+ hasTinfoil: !!tinfoilKey,
hasLocal: !!localAvailable,
});
- return !!(openaiKey || anthropicKey || geminiKey || groqKey || localAvailable);
+ return !!(openaiKey || anthropicKey || geminiKey || groqKey || tinfoilKey || localAvailable);
} catch (error) {
logger.logReasoning("API_KEY_CHECK_ERROR", {
error: (error as Error).message,
@@ -886,9 +904,13 @@ class ReasoningService extends BaseReasoningService {
if (provider !== "custom") {
this.apiKeyCache.delete(provider);
}
+ if (provider === "tinfoil") {
+ clearTinfoilClientCache();
+ }
logger.logReasoning("API_KEY_CACHE_CLEARED", { provider });
} else {
this.apiKeyCache.clear();
+ clearTinfoilClientCache();
logger.logReasoning("API_KEY_CACHE_CLEARED", { provider: "all" });
}
}
diff --git a/src/services/ai/inferenceProviders/index.ts b/src/services/ai/inferenceProviders/index.ts
index 58bfe5c46..b865f4a96 100644
--- a/src/services/ai/inferenceProviders/index.ts
+++ b/src/services/ai/inferenceProviders/index.ts
@@ -7,6 +7,7 @@ import { enterpriseProvider } from "./enterprise";
import { openwhisprProvider } from "./openwhispr";
import { lanProvider } from "./lan";
import { openaiProvider } from "./openai";
+import { tinfoilProvider } from "./tinfoil";
export const PROVIDER_REGISTRY: Readonly<Record<string, InferenceProvider>> = Object.freeze({
openai: openaiProvider,
@@ -14,6 +15,7 @@ export const PROVIDER_REGISTRY: Readonly<Record<string, InferenceProvider>> = Ob
anthropic: anthropicProvider,
gemini: geminiProvider,
groq: groqProvider,
+ tinfoil: tinfoilProvider,
local: localProvider,
bedrock: enterpriseProvider,
azure: enterpriseProvider,
diff --git a/src/services/ai/inferenceProviders/tinfoil.ts b/src/services/ai/inferenceProviders/tinfoil.ts
new file mode 100644
index 000000000..ab3682d37
--- /dev/null
+++ b/src/services/ai/inferenceProviders/tinfoil.ts
@@ -0,0 +1,87 @@
+import type { InferenceProvider } from "./types";
+import { TOKEN_LIMITS } from "../../../config/constants";
+import { getOpenAiApiConfig } from "../../../models/ModelRegistry";
+import { withRetry, createApiRetryStrategy } from "../../../utils/retry";
+import logger from "../../../utils/logger";
+import { applyThinkingSuppression } from "../thinkingSuppression";
+import { getTinfoilChatClient } from "../tinfoilClient";
+
+const REQUEST_TIMEOUT_MS = 30_000;
+
+export const tinfoilProvider: InferenceProvider = {
+ id: "tinfoil",
+ async call({ text, model, agentName, config, ctx }) {
+ logger.logReasoning("TINFOIL_START", { model, agentName });
+
+ const apiKey = await ctx.getApiKey("tinfoil");
+ // The client verifies enclave attestation before every request and
+ // refuses to send anything over an unverified transport.
+ const client = await getTinfoilChatClient(apiKey);
+
+ const systemPrompt = config.systemPrompt || ctx.getSystemPrompt(agentName);
+ const messages = [
+ { role: "system", content: systemPrompt },
+ { role: "user", content: text },
+ ];
+
+ const apiConfig = getOpenAiApiConfig(model);
+ const maxTokens =
+ config.maxTokens ||
+ Math.max(
+ 4096,
+ ctx.calculateMaxTokens(
+ text.length,
+ TOKEN_LIMITS.MIN_TOKENS,
+ TOKEN_LIMITS.MAX_TOKENS,
+ TOKEN_LIMITS.TOKEN_MULTIPLIER
+ )
+ );
+
+ const requestBody: Record<string, unknown> = {
+ model,
+ messages,
+ [apiConfig.tokenParam]: maxTokens,
+ };
+
+ if (apiConfig.supportsTemperature) {
+ requestBody.temperature = config.temperature ?? 0.3;
+ }
+
+ applyThinkingSuppression(requestBody, model, "tinfoil", config);
+
+ // 30s per attempt like sibling providers; SDK-internal retries off so
+ // withRetry stays the single retry layer.
+ const response = await withRetry(
+ () =>
+ client.chat.completions.create(requestBody as any, {
+ timeout: REQUEST_TIMEOUT_MS,
+ maxRetries: 0,
+ }),
+ createApiRetryStrategy()
+ );
+
+ const responseText =
+ response.choices
+ ?.map((choice: any) => choice?.message?.content)
+ .find((content: unknown) => typeof content === "string" && content.trim())
+ ?.trim() || "";
+
+ logger.logReasoning("TINFOIL_RESPONSE", {
+ model,
+ responseLength: responseText.length,
+ tokensUsed: response.usage?.total_tokens || 0,
+ success: true,
+ isEmpty: responseText.length === 0,
+ });
+
+ if (!responseText) {
+ logger.logReasoning("TINFOIL_EMPTY_RESPONSE_FALLBACK", {
+ model,
+ originalTextLength: text.length,
+ });
+ return text;
+ }
+
+ return responseText;
+ },
+};
diff --git a/src/services/ai/providers.ts b/src/services/ai/providers.ts
index f49da8884..d9ceecda1 100644
--- a/src/services/ai/providers.ts
+++ b/src/services/ai/providers.ts
@@ -3,6 +3,7 @@ import { createGroq } from "@ai-sdk/groq";
import { createAnthropic } from "@ai-sdk/anthropic";
import { createGoogleGenerativeAI } from "@ai-sdk/google";
import type { LanguageModel } from "ai";
+import { getTinfoilLanguageModel } from "./tinfoilClient";
// Renderer-side AI SDK factory. Cloud + local only — enterprise providers
// (bedrock/azure/vertex) run in the main process via the
@@ -10,12 +11,12 @@ import type { LanguageModel } from "ai";
// APIs (fs, process, AWS credential chain) that don't work in the browser.
// See `src/helpers/enterpriseAiProviders.js` for the main-process counterpart.
-export function getAIModel(
+export async function getAIModel(
provider: string,
model: string,
apiKey: string,
baseURL?: string
-): LanguageModel {
+): Promise<LanguageModel> {
switch (provider) {
case "openai":
return createOpenAI({ apiKey })(model);
@@ -25,6 +26,8 @@ export function getAIModel(
return createAnthropic({ apiKey })(model);
case "gemini":
return createGoogleGenerativeAI({ apiKey })(model);
+ case "tinfoil":
+ return getTinfoilLanguageModel(apiKey, model);
case "custom":
return createOpenAI({ apiKey, baseURL })(model);
case "local":
diff --git a/src/services/ai/tinfoilClient.ts b/src/services/ai/tinfoilClient.ts
new file mode 100644
index 000000000..670963a02
--- /dev/null
+++ b/src/services/ai/tinfoilClient.ts
@@ -0,0 +1,77 @@
+import type { LanguageModel } from "ai";
+import type { TinfoilAI } from "tinfoil";
+
+type TinfoilModule = typeof import("tinfoil");
+type TinfoilAISDKProvider = Awaited<ReturnType<TinfoilModule["createTinfoilAI"]>>;
+
+let tinfoilModulePromise: Promise<TinfoilModule> | null = null;
+
+const chatClientCache = new Map<string, Promise<TinfoilAI>>();
+const aiSdkProviderCache = new Map<string, Promise<TinfoilAISDKProvider>>();
+
+function loadTinfoil(): Promise<TinfoilModule> {
+ if (!tinfoilModulePromise) {
+ // Don't cache a failed import — the next call should retry.
+ tinfoilModulePromise = import("tinfoil").catch((error) => {
+ tinfoilModulePromise = null;
+ throw error;
+ });
+ }
+ return tinfoilModulePromise;
+}
+
+function normalizeApiKey(apiKey: string): string {
+ const key = apiKey?.trim() || "";
+ if (!key) {
+ throw new Error("Tinfoil API key not configured");
+ }
+ return key;
+}
+
+export async function getTinfoilChatClient(apiKey: string): Promise<TinfoilAI> {
+ const key = normalizeApiKey(apiKey);
+ const cached = chatClientCache.get(key);
+ if (cached) return cached;
+
+ const clientPromise = loadTinfoil()
+ .then(({ TinfoilAI }) => {
+ return new TinfoilAI({
+ apiKey: key,
+ dangerouslyAllowBrowser: true,
+ });
+ })
+ .catch((error) => {
+ chatClientCache.delete(key);
+ throw error;
+ });
+ chatClientCache.set(key, clientPromise);
+ return clientPromise;
+}
+
+async function getTinfoilAISDKProvider(apiKey: string): Promise<TinfoilAISDKProvider> {
+ const key = normalizeApiKey(apiKey);
+ const cached = aiSdkProviderCache.get(key);
+ if (cached) return cached;
+
+ const providerPromise = loadTinfoil()
+ .then(({ createTinfoilAI }) => createTinfoilAI(key))
+ .catch((error) => {
+ aiSdkProviderCache.delete(key);
+ throw error;
+ });
+ aiSdkProviderCache.set(key, providerPromise);
+ return providerPromise;
+}
+
+export async function getTinfoilLanguageModel(
+ apiKey: string,
+ model: string
+): Promise<LanguageModel> {
+ const provider = await getTinfoilAISDKProvider(apiKey);
+ return provider(model);
+}
+
+export function clearTinfoilClientCache(): void {
+ chatClientCache.clear();
+ aiSdkProviderCache.clear();
+}