From 8ff004f290c5e85241057bc545162d07e0a2d063 Mon Sep 17 00:00:00 2001
From: Mauro Druwel <mauro.druwel@gmail.com>
Date: Thu, 19 Feb 2026 14:52:25 +0100
Subject: [PATCH 1/5] feat: expand LLM provider support to include OpenRouter
 and NVIDIA NIM

---
 README.md                   | 16 ++++++++--------
 docs/ARCHITECTURE.md        | 27 ++++++++++++++++++---------
 main/cli/serial_cli.c       |  2 +-
 main/llm/llm_proxy.c        | 30 +++++++++++++++++++++++++-----
 main/llm/llm_proxy.h        |  2 +-
 main/mimi_config.h          |  2 ++
 main/mimi_secrets.h.example |  2 +-
 7 files changed, 56 insertions(+), 25 deletions(-)

diff --git a/README.md b/README.md
index cead1927..e48d318c 100644
--- a/README.md
+++ b/README.md
@@ -31,7 +31,7 @@ MimiClaw turns a tiny ESP32-S3 board into a personal AI assistant. Plug it into
 
 ![](assets/mimiclaw.png)
 
-You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it into an agent loop — the LLM thinks, calls tools, reads memory — and sends the reply back. Supports both **Anthropic (Claude)** and **OpenAI (GPT)** as providers, switchable at runtime. Everything runs on a single $5 chip with all your data stored locally on flash.
+You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it into an agent loop — the LLM thinks, calls tools, reads memory — and sends the reply back. Supports **Anthropic (Claude)**, **OpenAI (GPT)**, **OpenRouter**, and **NVIDIA NIM (completely free)** as providers, switchable at runtime. Everything runs on a single $5 chip with all your data stored locally on flash.
 
 ## Quick Start
 
@@ -40,7 +40,7 @@ You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it int
 - An **ESP32-S3 dev board** with 16 MB flash and 8 MB PSRAM (e.g. Xiaozhi AI board, ~$10)
 - A **USB Type-C cable**
 - A **Telegram bot token** — talk to [@BotFather](https://t.me/BotFather) on Telegram to create one
-- An **Anthropic API key** — from [console.anthropic.com](https://console.anthropic.com), or an **OpenAI API key** — from [platform.openai.com](https://platform.openai.com)
+- An **API key** — from [Anthropic](https://console.anthropic.com), [OpenAI](https://platform.openai.com), [OpenRouter](https://openrouter.ai), or [NVIDIA NIM (free)](https://org.ngc.nvidia.com/setup/api-keys)
 
 ### Install
 
@@ -69,7 +69,7 @@ Edit `main/mimi_secrets.h`:
 #define MIMI_SECRET_WIFI_PASS       "YourWiFiPassword"
 #define MIMI_SECRET_TG_TOKEN        "123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11"
 #define MIMI_SECRET_API_KEY         "sk-ant-api03-xxxxx"
-#define MIMI_SECRET_MODEL_PROVIDER  "anthropic"     // "anthropic" or "openai"
+#define MIMI_SECRET_MODEL_PROVIDER  "anthropic"     // "anthropic", "openai", "openrouter", or "nvidia_nim"
 #define MIMI_SECRET_SEARCH_KEY      ""              // optional: Brave Search API key
 #define MIMI_SECRET_PROXY_HOST      ""              // optional: e.g. "10.0.0.1"
 #define MIMI_SECRET_PROXY_PORT      ""              // optional: e.g. "7897"
@@ -108,8 +108,8 @@ Connect via serial to configure or debug. **Config commands** let you change set
 ```
 mimi> wifi_set MySSID MyPassword   # change WiFi network
 mimi> set_tg_token 123456:ABC...   # change Telegram bot token
-mimi> set_api_key sk-ant-api03-... # change API key (Anthropic or OpenAI)
-mimi> set_model_provider openai    # switch provider (anthropic|openai)
+mimi> set_api_key sk-ant-api03-... # change API key (Anthropic, OpenAI, OpenRouter, or NVIDIA NIM)
+mimi> set_model_provider openai    # switch provider (anthropic|openai|openrouter|nvidia_nim)
 mimi> set_model gpt-4o             # change LLM model
 mimi> set_proxy 127.0.0.1 7897  # set HTTP proxy
 mimi> clear_proxy                  # remove proxy
@@ -148,7 +148,7 @@ MimiClaw stores everything as plain text files you can read and edit:
 
 ## Tools
 
-MimiClaw supports tool calling for both Anthropic and OpenAI — the LLM can call tools during a conversation and loop until the task is done (ReAct pattern).
+MimiClaw supports tool calling for Anthropic, OpenAI, OpenRouter, and NVIDIA NIM — the LLM can call tools during a conversation and loop until the task is done (ReAct pattern).
 
 | Tool | Description |
 |------|-------------|
@@ -178,10 +178,10 @@ This turns MimiClaw into a proactive assistant — write tasks to `HEARTBEAT.md`
 - **OTA updates** — flash new firmware over WiFi, no USB needed
 - **Dual-core** — network I/O and AI processing run on separate CPU cores
 - **HTTP proxy** — CONNECT tunnel support for restricted networks
-- **Multi-provider** — supports both Anthropic (Claude) and OpenAI (GPT), switchable at runtime
+- **Multi-provider** — supports Anthropic (Claude), OpenAI (GPT), OpenRouter, and NVIDIA NIM (completely free), switchable at runtime
 - **Cron scheduler** — the AI can schedule its own recurring and one-shot tasks, persisted across reboots
 - **Heartbeat** — periodically checks a task file and prompts the AI to act autonomously
-- **Tool use** — ReAct agent loop with tool calling for both providers
+- **Tool use** — ReAct agent loop with tool calling for all providers
 
 ## For Developers
 
diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
index b8b399a4..c70cecfa 100644
--- a/docs/ARCHITECTURE.md
+++ b/docs/ARCHITECTURE.md
@@ -54,12 +54,13 @@ Telegram App (User)
 │   └──────────────────────────────────────────┘    │
 └───────────────────────────────────────────────────┘
          │
-         │  Anthropic Messages API (HTTPS)
+         │  Anthropic/OpenAI/OpenRouter API (HTTPS)
          │  + Brave Search API (HTTPS)
          ▼
-   ┌───────────┐   ┌──────────────┐
-   │ Claude API │   │ Brave Search │
-   └───────────┘   └──────────────┘
+   ┌──────────────┐   ┌──────────────┐
+   │ LLM Provider │   │ Brave Search │
+   │ (API Gateway)│   │              │
+   └──────────────┘   └──────────────┘
 ```
 
 ---
@@ -75,7 +76,7 @@ Telegram App (User)
    b. Build system prompt (SOUL.md + USER.md + MEMORY.md + recent notes + tool guidance)
    c. Build cJSON messages array (history + current message)
    d. ReAct loop (max 10 iterations):
-      i.   Call Claude API via HTTPS (non-streaming, with tools array)
+      i.   Call LLM provider API via HTTPS (non-streaming, with tools array)
       ii.  Parse JSON response → text blocks + tool_use blocks
       iii. If stop_reason == "tool_use":
            - Execute each tool (e.g. web_search → Brave Search API)
@@ -158,7 +159,7 @@ main/
 | Task               | Core | Priority | Stack  | Description                          |
 |--------------------|------|----------|--------|--------------------------------------|
 | `tg_poll`          | 0    | 5        | 12 KB  | Telegram long polling (30s timeout)  |
-| `agent_loop`       | 1    | 6        | 12 KB  | Message processing + Claude API call |
+| `agent_loop`       | 1    | 6        | 12 KB  | Message processing + LLM provider API call |
 | `outbound`         | 0    | 5        | 8 KB   | Route responses to Telegram / WS     |
 | `serial_cli`       | 0    | 3        | 4 KB   | USB serial console REPL              |
 | httpd (internal)   | 0    | 5        | —      | WebSocket server (esp_http_server)   |
@@ -232,7 +233,7 @@ All configuration is done exclusively through `mimi_secrets.h` at build time. Th
 | `MIMI_SECRET_WIFI_SSID`     | WiFi SSID                               |
 | `MIMI_SECRET_WIFI_PASS`     | WiFi password                           |
 | `MIMI_SECRET_TG_TOKEN`      | Telegram Bot API token                  |
-| `MIMI_SECRET_API_KEY`       | Anthropic API key                       |
+| `MIMI_SECRET_API_KEY`       | LLM API key (Anthropic, OpenAI, OpenRouter, or NVIDIA NIM) |
 | `MIMI_SECRET_MODEL`         | Model ID (default: claude-opus-4-6)     |
 | `MIMI_SECRET_PROXY_HOST`    | HTTP proxy hostname/IP (optional)       |
 | `MIMI_SECRET_PROXY_PORT`    | HTTP proxy port (optional)              |
@@ -278,7 +279,15 @@ Client `chat_id` is auto-assigned on connection (`ws_<fd>`) but can be overridde
 
 ---
 
-## Claude API Integration
+## LLM Provider Integration
+
+MimiClaw supports multiple LLM providers via a unified API proxy:
+- **Anthropic** (`anthropic`) - Native Messages API
+- **OpenAI** (`openai`) - Chat Completions API
+- **OpenRouter** (`openrouter`) - OpenAI-compatible endpoint
+- **NVIDIA NIM** (`nvidia_nim`) - OpenAI-compatible endpoint, **completely free**
+
+### Anthropic API Format
 
 Endpoint: `POST https://api.anthropic.com/v1/messages`
 
@@ -388,7 +397,7 @@ The CLI provides debug and maintenance commands only. All configuration is done
 | `session/manager.py`        | `memory/session_mgr.c`         | JSONL per chat, ring buffer  |
 | `channels/telegram.py`      | `telegram/telegram_bot.c`      | Raw HTTP, no python-telegram-bot |
 | `bus/events.py` + `queue.py`| `bus/message_bus.c`            | FreeRTOS queues vs asyncio   |
-| `providers/litellm_provider.py` | `llm/llm_proxy.c`         | Direct Anthropic API only    |
+| `providers/litellm_provider.py` | `llm/llm_proxy.c`         | Multi-provider LLM API support  |
 | `config/schema.py`          | `mimi_config.h` + `mimi_secrets.h` | Build-time secrets only  |
 | `cli/commands.py`           | `cli/serial_cli.c`             | esp_console REPL             |
 | `agent/tools/*`             | `tools/tool_registry.c` + `tool_web_search.c` | web_search via Brave API |
diff --git a/main/cli/serial_cli.c b/main/cli/serial_cli.c
index d2315c4f..6e369a7c 100644
--- a/main/cli/serial_cli.c
+++ b/main/cli/serial_cli.c
@@ -623,7 +623,7 @@ esp_err_t serial_cli_init(void)
     esp_console_cmd_register(&model_cmd);
 
     /* set_model_provider */
-    provider_args.provider = arg_str1(NULL, NULL, "<provider>", "Model provider (anthropic|openai)");
+    provider_args.provider = arg_str1(NULL, NULL, "<provider>", "Model provider (anthropic|openai|openrouter|nvidia_nim)");
     provider_args.end = arg_end(1);
     esp_console_cmd_t provider_cmd = {
         .command = "set_model_provider",
diff --git a/main/llm/llm_proxy.c b/main/llm/llm_proxy.c
index dba29610..250929c5 100644
--- a/main/llm/llm_proxy.c
+++ b/main/llm/llm_proxy.c
@@ -139,19 +139,39 @@ static bool provider_is_openai(void)
     return strcmp(s_provider, "openai") == 0;
 }
 
+static bool provider_is_openrouter(void)
+{
+    return strcmp(s_provider, "openrouter") == 0;
+}
+
+static bool provider_is_nvidia_nim(void)
+{
+    return strcmp(s_provider, "nvidia_nim") == 0;
+}
+
+/** Return the full API URL for the active provider. */
 static const char *llm_api_url(void)
 {
-    return provider_is_openai() ? MIMI_OPENAI_API_URL : MIMI_LLM_API_URL;
+    if (provider_is_openai()) return MIMI_OPENAI_API_URL;
+    if (provider_is_openrouter()) return MIMI_OPENROUTER_API_URL;
+    if (provider_is_nvidia_nim()) return MIMI_NVIDIA_NIM_API_URL;
+    return MIMI_LLM_API_URL;
 }
 
+/** Return the HTTP Host header value for the active provider. */
 static const char *llm_api_host(void)
 {
-    return provider_is_openai() ? "api.openai.com" : "api.anthropic.com";
+    if (provider_is_openai()) return "api.openai.com";
+    if (provider_is_openrouter()) return "openrouter.ai";
+    if (provider_is_nvidia_nim()) return "integrate.api.nvidia.com";
+    return "api.anthropic.com";
 }
 
+/** Return the HTTP request path for the active provider. */
 static const char *llm_api_path(void)
 {
-    return provider_is_openai() ? "/v1/chat/completions" : "/v1/messages";
+    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia_nim()) return "/v1/chat/completions";
+    return "/v1/messages";
 }
 
 /* ── Init ─────────────────────────────────────────────────────── */
@@ -217,7 +237,7 @@ static esp_err_t llm_http_direct(const char *post_data, resp_buf_t *rb, int *out
 
     esp_http_client_set_method(client, HTTP_METHOD_POST);
     esp_http_client_set_header(client, "Content-Type", "application/json");
-    if (provider_is_openai()) {
+    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia_nim()) {
         if (s_api_key[0]) {
             char auth[LLM_API_KEY_MAX_LEN + 16];
             snprintf(auth, sizeof(auth), "Bearer %s", s_api_key);
@@ -245,7 +265,7 @@ static esp_err_t llm_http_via_proxy(const char *post_data, resp_buf_t *rb, int *
     int body_len = strlen(post_data);
     char header[1024];
     int hlen = 0;
-    if (provider_is_openai()) {
+    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia_nim()) {
         hlen = snprintf(header, sizeof(header),
             "POST %s HTTP/1.1\r\n"
             "Host: %s\r\n"
diff --git a/main/llm/llm_proxy.h b/main/llm/llm_proxy.h
index 03c3967d..cd184379 100644
--- a/main/llm/llm_proxy.h
+++ b/main/llm/llm_proxy.h
@@ -18,7 +18,7 @@ esp_err_t llm_proxy_init(void);
 esp_err_t llm_set_api_key(const char *api_key);
 
 /**
- * Save the LLM provider to NVS. (e.g. "anthropic", "openai")
+ * Save the LLM provider to NVS. (e.g. "anthropic", "openai", "openrouter", "nvidia_nim")
  */
 esp_err_t llm_set_provider(const char *provider);
 
diff --git a/main/mimi_config.h b/main/mimi_config.h
index 47a35973..b033cceb 100644
--- a/main/mimi_config.h
+++ b/main/mimi_config.h
@@ -67,6 +67,8 @@
 #define MIMI_LLM_MAX_TOKENS          4096
 #define MIMI_LLM_API_URL             "https://api.anthropic.com/v1/messages"
 #define MIMI_OPENAI_API_URL          "https://api.openai.com/v1/chat/completions"
+#define MIMI_OPENROUTER_API_URL      "https://openrouter.ai/api/v1/chat/completions"
+#define MIMI_NVIDIA_NIM_API_URL      "https://integrate.api.nvidia.com/v1/chat/completions"
 #define MIMI_LLM_API_VERSION         "2023-06-01"
 #define MIMI_LLM_STREAM_BUF_SIZE     (32 * 1024)
 #define MIMI_LLM_LOG_VERBOSE_PAYLOAD 0
diff --git a/main/mimi_secrets.h.example b/main/mimi_secrets.h.example
index bf76defb..9f321d02 100644
--- a/main/mimi_secrets.h.example
+++ b/main/mimi_secrets.h.example
@@ -17,7 +17,7 @@
 /* Telegram Bot */
 #define MIMI_SECRET_TG_TOKEN        ""
 
-/* Anthropic API */
+/* LLM API (anthropic, openai, openrouter, or nvidia_nim)
 #define MIMI_SECRET_API_KEY         ""
 #define MIMI_SECRET_MODEL           ""
 #define MIMI_SECRET_MODEL_PROVIDER  "anthropic"

From eb545ab4acdd2eef35589713ad87764f88c9926a Mon Sep 17 00:00:00 2001
From: Mauro Druwel <46003176+MauroDruwel@users.noreply.github.com>
Date: Fri, 20 Feb 2026 10:56:42 +0000
Subject: [PATCH 2/5] Fix pr comments

---
 README.md                   |  6 +++---
 docs/ARCHITECTURE.md        |  4 ++--
 main/cli/serial_cli.c       |  2 +-
 main/llm/llm_proxy.c        | 33 ++++++++++++++++++++-------------
 main/llm/llm_proxy.h        |  2 +-
 main/mimi_secrets.h.example |  2 +-
 6 files changed, 28 insertions(+), 21 deletions(-)

diff --git a/README.md b/README.md
index e48d318c..69e3065c 100644
--- a/README.md
+++ b/README.md
@@ -31,7 +31,7 @@ MimiClaw turns a tiny ESP32-S3 board into a personal AI assistant. Plug it into
 
 ![](assets/mimiclaw.png)
 
-You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it into an agent loop — the LLM thinks, calls tools, reads memory — and sends the reply back. Supports **Anthropic (Claude)**, **OpenAI (GPT)**, **OpenRouter**, and **NVIDIA NIM (completely free)** as providers, switchable at runtime. Everything runs on a single $5 chip with all your data stored locally on flash.
+You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it into an agent loop — the LLM thinks, calls tools, reads memory — and sends the reply back. Supports **Anthropic (Claude)**, **OpenAI (GPT)**, **OpenRouter**, and **[NVIDIA NIM](https://www.nvidia.com/en-us/ai/#nim) (free for development for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing)** as providers, switchable at runtime. Everything runs on a single $5 chip with all your data stored locally on flash.
 
 ## Quick Start
 
@@ -69,7 +69,7 @@ Edit `main/mimi_secrets.h`:
 #define MIMI_SECRET_WIFI_PASS       "YourWiFiPassword"
 #define MIMI_SECRET_TG_TOKEN        "123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11"
 #define MIMI_SECRET_API_KEY         "sk-ant-api03-xxxxx"
-#define MIMI_SECRET_MODEL_PROVIDER  "anthropic"     // "anthropic", "openai", "openrouter", or "nvidia_nim"
+#define MIMI_SECRET_MODEL_PROVIDER  "anthropic"     // "anthropic", "openai", "openrouter", or "nvidia"
 #define MIMI_SECRET_SEARCH_KEY      ""              // optional: Brave Search API key
 #define MIMI_SECRET_PROXY_HOST      ""              // optional: e.g. "10.0.0.1"
 #define MIMI_SECRET_PROXY_PORT      ""              // optional: e.g. "7897"
@@ -109,7 +109,7 @@ Connect via serial to configure or debug. **Config commands** let you change set
 mimi> wifi_set MySSID MyPassword   # change WiFi network
 mimi> set_tg_token 123456:ABC...   # change Telegram bot token
 mimi> set_api_key sk-ant-api03-... # change API key (Anthropic, OpenAI, OpenRouter, or NVIDIA NIM)
-mimi> set_model_provider openai    # switch provider (anthropic|openai|openrouter|nvidia_nim)
+mimi> set_model_provider openai    # switch provider (anthropic|openai|openrouter|nvidia)
 mimi> set_model gpt-4o             # change LLM model
 mimi> set_proxy 127.0.0.1 7897  # set HTTP proxy
 mimi> clear_proxy                  # remove proxy
diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
index c70cecfa..b8302654 100644
--- a/docs/ARCHITECTURE.md
+++ b/docs/ARCHITECTURE.md
@@ -54,7 +54,7 @@ Telegram App (User)
 │   └──────────────────────────────────────────┘    │
 └───────────────────────────────────────────────────┘
          │
-         │  Anthropic/OpenAI/OpenRouter API (HTTPS)
+         │  Anthropic/OpenAI/OpenRouter/NVIDIA NIM API (HTTPS)
          │  + Brave Search API (HTTPS)
          ▼
    ┌──────────────┐   ┌──────────────┐
@@ -285,7 +285,7 @@ MimiClaw supports multiple LLM providers via a unified API proxy:
 - **Anthropic** (`anthropic`) - Native Messages API
 - **OpenAI** (`openai`) - Chat Completions API
 - **OpenRouter** (`openrouter`) - OpenAI-compatible endpoint
-- **NVIDIA NIM** (`nvidia_nim`) - OpenAI-compatible endpoint, **completely free**
+- **NVIDIA NIM** (`nvidia`) - OpenAI-compatible endpoint (free for development/prototyping for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing)
 
 ### Anthropic API Format
 
diff --git a/main/cli/serial_cli.c b/main/cli/serial_cli.c
index 6e369a7c..3108b4e2 100644
--- a/main/cli/serial_cli.c
+++ b/main/cli/serial_cli.c
@@ -623,7 +623,7 @@ esp_err_t serial_cli_init(void)
     esp_console_cmd_register(&model_cmd);
 
     /* set_model_provider */
-    provider_args.provider = arg_str1(NULL, NULL, "<provider>", "Model provider (anthropic|openai|openrouter|nvidia_nim)");
+    provider_args.provider = arg_str1(NULL, NULL, "<provider>", "Model provider (anthropic|openai|openrouter|nvidia)");
     provider_args.end = arg_end(1);
     esp_console_cmd_t provider_cmd = {
         .command = "set_model_provider",
diff --git a/main/llm/llm_proxy.c b/main/llm/llm_proxy.c
index 250929c5..b7189590 100644
--- a/main/llm/llm_proxy.c
+++ b/main/llm/llm_proxy.c
@@ -144,9 +144,9 @@ static bool provider_is_openrouter(void)
     return strcmp(s_provider, "openrouter") == 0;
 }
 
-static bool provider_is_nvidia_nim(void)
+static bool provider_is_nvidia(void)
 {
-    return strcmp(s_provider, "nvidia_nim") == 0;
+    return strcmp(s_provider, "nvidia") == 0;
 }
 
 /** Return the full API URL for the active provider. */
@@ -154,7 +154,7 @@ static const char *llm_api_url(void)
 {
     if (provider_is_openai()) return MIMI_OPENAI_API_URL;
     if (provider_is_openrouter()) return MIMI_OPENROUTER_API_URL;
-    if (provider_is_nvidia_nim()) return MIMI_NVIDIA_NIM_API_URL;
+    if (provider_is_nvidia()) return MIMI_NVIDIA_NIM_API_URL;
     return MIMI_LLM_API_URL;
 }
 
@@ -163,17 +163,24 @@ static const char *llm_api_host(void)
 {
     if (provider_is_openai()) return "api.openai.com";
     if (provider_is_openrouter()) return "openrouter.ai";
-    if (provider_is_nvidia_nim()) return "integrate.api.nvidia.com";
+    if (provider_is_nvidia()) return "integrate.api.nvidia.com";
     return "api.anthropic.com";
 }
 
 /** Return the HTTP request path for the active provider. */
 static const char *llm_api_path(void)
 {
-    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia_nim()) return "/v1/chat/completions";
+    if (provider_is_openrouter()) return "/api/v1/chat/completions";
+    if (provider_is_openai() || provider_is_nvidia()) return "/v1/chat/completions";
     return "/v1/messages";
 }
 
+/** Return true for OpenAI-compatible providers (OpenAI, OpenRouter, NVIDIA NIM). */
+static bool provider_is_openai_compat(void)
+{
+    return provider_is_openai() || provider_is_openrouter() || provider_is_nvidia();
+}
+
 /* ── Init ─────────────────────────────────────────────────────── */
 
 esp_err_t llm_proxy_init(void)
@@ -237,7 +244,7 @@ static esp_err_t llm_http_direct(const char *post_data, resp_buf_t *rb, int *out
 
     esp_http_client_set_method(client, HTTP_METHOD_POST);
     esp_http_client_set_header(client, "Content-Type", "application/json");
-    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia_nim()) {
+    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) {
         if (s_api_key[0]) {
             char auth[LLM_API_KEY_MAX_LEN + 16];
             snprintf(auth, sizeof(auth), "Bearer %s", s_api_key);
@@ -265,7 +272,7 @@ static esp_err_t llm_http_via_proxy(const char *post_data, resp_buf_t *rb, int *
     int body_len = strlen(post_data);
     char header[1024];
     int hlen = 0;
-    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia_nim()) {
+    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) {
         hlen = snprintf(header, sizeof(header),
             "POST %s HTTP/1.1\r\n"
             "Host: %s\r\n"
@@ -552,13 +559,13 @@ esp_err_t llm_chat(const char *system_prompt, const char *messages_json,
     /* Build request body (non-streaming) */
     cJSON *body = cJSON_CreateObject();
     cJSON_AddStringToObject(body, "model", s_model);
-    if (provider_is_openai()) {
+    if (provider_is_openai_compat()) {
         cJSON_AddNumberToObject(body, "max_completion_tokens", MIMI_LLM_MAX_TOKENS);
     } else {
         cJSON_AddNumberToObject(body, "max_tokens", MIMI_LLM_MAX_TOKENS);
     }
 
-    if (provider_is_openai()) {
+    if (provider_is_openai_compat()) {
         cJSON *messages = cJSON_Parse(messages_json);
         if (!messages) {
             messages = cJSON_CreateArray();
@@ -635,7 +642,7 @@ esp_err_t llm_chat(const char *system_prompt, const char *messages_json,
         return ESP_FAIL;
     }
 
-    if (provider_is_openai()) {
+    if (provider_is_openai_compat()) {
         extract_text_openai(root, response_buf, buf_size);
     } else {
         extract_text_anthropic(root, response_buf, buf_size);
@@ -678,13 +685,13 @@ esp_err_t llm_chat_tools(const char *system_prompt,
     /* Build request body (non-streaming) */
     cJSON *body = cJSON_CreateObject();
     cJSON_AddStringToObject(body, "model", s_model);
-    if (provider_is_openai()) {
+    if (provider_is_openai_compat()) {
         cJSON_AddNumberToObject(body, "max_completion_tokens", MIMI_LLM_MAX_TOKENS);
     } else {
         cJSON_AddNumberToObject(body, "max_tokens", MIMI_LLM_MAX_TOKENS);
     }
 
-    if (provider_is_openai()) {
+    if (provider_is_openai_compat()) {
         cJSON *openai_msgs = convert_messages_openai(system_prompt, messages);
         cJSON_AddItemToObject(body, "messages", openai_msgs);
 
@@ -754,7 +761,7 @@ esp_err_t llm_chat_tools(const char *system_prompt,
         return ESP_FAIL;
     }
 
-    if (provider_is_openai()) {
+    if (provider_is_openai_compat()) {
         cJSON *choices = cJSON_GetObjectItem(root, "choices");
         cJSON *choice0 = choices && cJSON_IsArray(choices) ? cJSON_GetArrayItem(choices, 0) : NULL;
         if (choice0) {
diff --git a/main/llm/llm_proxy.h b/main/llm/llm_proxy.h
index cd184379..3fd183e9 100644
--- a/main/llm/llm_proxy.h
+++ b/main/llm/llm_proxy.h
@@ -18,7 +18,7 @@ esp_err_t llm_proxy_init(void);
 esp_err_t llm_set_api_key(const char *api_key);
 
 /**
- * Save the LLM provider to NVS. (e.g. "anthropic", "openai", "openrouter", "nvidia_nim")
+ * Save the LLM provider to NVS. (e.g. "anthropic", "openai", "openrouter", "nvidia")
  */
 esp_err_t llm_set_provider(const char *provider);
 
diff --git a/main/mimi_secrets.h.example b/main/mimi_secrets.h.example
index 9f321d02..d60d49e6 100644
--- a/main/mimi_secrets.h.example
+++ b/main/mimi_secrets.h.example
@@ -17,7 +17,7 @@
 /* Telegram Bot */
 #define MIMI_SECRET_TG_TOKEN        ""
 
-/* LLM API (anthropic, openai, openrouter, or nvidia_nim)
+/* LLM API (anthropic, openai, openrouter, or nvidia) */
 #define MIMI_SECRET_API_KEY         ""
 #define MIMI_SECRET_MODEL           ""
 #define MIMI_SECRET_MODEL_PROVIDER  "anthropic"

From 9c9d72043b964694d84d123433f7ebe5666aa3fc Mon Sep 17 00:00:00 2001
From: Mauro Druwel <46003176+MauroDruwel@users.noreply.github.com>
Date: Fri, 20 Feb 2026 11:08:11 +0000
Subject: [PATCH 3/5] Update readme with pr comments

---
 README.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index 69e3065c..199b8da3 100644
--- a/README.md
+++ b/README.md
@@ -40,7 +40,7 @@ You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it int
 - An **ESP32-S3 dev board** with 16 MB flash and 8 MB PSRAM (e.g. Xiaozhi AI board, ~$10)
 - A **USB Type-C cable**
 - A **Telegram bot token** — talk to [@BotFather](https://t.me/BotFather) on Telegram to create one
-- An **API key** — from [Anthropic](https://console.anthropic.com), [OpenAI](https://platform.openai.com), [OpenRouter](https://openrouter.ai), or [NVIDIA NIM (free)](https://org.ngc.nvidia.com/setup/api-keys)
+- An **API key** — from [Anthropic](https://console.anthropic.com), [OpenAI](https://platform.openai.com), [OpenRouter](https://openrouter.ai), or [NVIDIA NIM](https://org.ngc.nvidia.com/setup/api-keys) (free for development for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing)
 
 ### Install
 
@@ -178,7 +178,7 @@ This turns MimiClaw into a proactive assistant — write tasks to `HEARTBEAT.md`
 - **OTA updates** — flash new firmware over WiFi, no USB needed
 - **Dual-core** — network I/O and AI processing run on separate CPU cores
 - **HTTP proxy** — CONNECT tunnel support for restricted networks
-- **Multi-provider** — supports Anthropic (Claude), OpenAI (GPT), OpenRouter, and NVIDIA NIM (completely free), switchable at runtime
+- **Multi-provider** — supports Anthropic (Claude), OpenAI (GPT), OpenRouter, and NVIDIA NIM (free for development for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing), switchable at runtime
 - **Cron scheduler** — the AI can schedule its own recurring and one-shot tasks, persisted across reboots
 - **Heartbeat** — periodically checks a task file and prompts the AI to act autonomously
 - **Tool use** — ReAct agent loop with tool calling for all providers

From f57343b373da01edecf13963da3381928b0cf29a Mon Sep 17 00:00:00 2001
From: Mauro Druwel <mauro.druwel@gmail.com>
Date: Sun, 15 Mar 2026 10:37:37 +0100
Subject: [PATCH 4/5] docs: update README and ARCHITECTURE for NVIDIA NIM feat:
 add TODO in llm_proxy.c for deriving host from configured API URL

---
 README.md            | 6 +++---
 docs/ARCHITECTURE.md | 2 +-
 main/llm/llm_proxy.c | 1 +
 3 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/README.md b/README.md
index ec468181..c1fb379e 100644
--- a/README.md
+++ b/README.md
@@ -31,7 +31,7 @@ MimiClaw turns a tiny ESP32-S3 board into a personal AI assistant. Plug it into
 
 ![](assets/mimiclaw.png)
 
-You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it into an agent loop — the LLM thinks, calls tools, reads memory — and sends the reply back. Supports **Anthropic (Claude)**, **OpenAI (GPT)**, **OpenRouter**, and **[NVIDIA NIM](https://www.nvidia.com/en-us/ai/#nim) (free for development for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing)** as providers, switchable at runtime. Everything runs on a single $5 chip with all your data stored locally on flash.
+You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it into an agent loop — the LLM thinks, calls tools, reads memory — and sends the reply back. Supports **Anthropic (Claude)**, **OpenAI (GPT)**, **OpenRouter**, and **NVIDIA NIM** as providers, switchable at runtime. Everything runs on a single $5 chip with all your data stored locally on flash.
 
 ## Quick Start
 
@@ -40,7 +40,7 @@ You send a message on Telegram. The ESP32-S3 picks it up over WiFi, feeds it int
 - An **ESP32-S3 dev board** with 16 MB flash and 8 MB PSRAM (e.g. Xiaozhi AI board, ~$10)
 - A **USB Type-C cable**
 - A **Telegram bot token** — talk to [@BotFather](https://t.me/BotFather) on Telegram to create one
-- An **API key** — from [Anthropic](https://console.anthropic.com), [OpenAI](https://platform.openai.com), [OpenRouter](https://openrouter.ai), or [NVIDIA NIM](https://org.ngc.nvidia.com/setup/api-keys) (free for development for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing)
+- An **API key** — from [Anthropic](https://console.anthropic.com), [OpenAI](https://platform.openai.com), [OpenRouter](https://openrouter.ai), or [NVIDIA NIM](https://org.ngc.nvidia.com/setup/api-keys)
 
 ### Install
 
@@ -237,7 +237,7 @@ This turns MimiClaw into a proactive assistant — write tasks to `HEARTBEAT.md`
 - **OTA updates** — flash new firmware over WiFi, no USB needed
 - **Dual-core** — network I/O and AI processing run on separate CPU cores
 - **HTTP proxy** — CONNECT tunnel support for restricted networks
-- **Multi-provider** — supports Anthropic (Claude), OpenAI (GPT), OpenRouter, and NVIDIA NIM (free for development for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing), switchable at runtime
+- **Multi-provider** — supports Anthropic (Claude), OpenAI (GPT), OpenRouter, and NVIDIA NIM, switchable at runtime
 - **Cron scheduler** — the AI can schedule its own recurring and one-shot tasks, persisted across reboots
 - **Heartbeat** — periodically checks a task file and prompts the AI to act autonomously
 - **Tool use** — ReAct agent loop with tool calling for all providers
diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
index b8302654..3387741d 100644
--- a/docs/ARCHITECTURE.md
+++ b/docs/ARCHITECTURE.md
@@ -285,7 +285,7 @@ MimiClaw supports multiple LLM providers via a unified API proxy:
 - **Anthropic** (`anthropic`) - Native Messages API
 - **OpenAI** (`openai`) - Chat Completions API
 - **OpenRouter** (`openrouter`) - OpenAI-compatible endpoint
-- **NVIDIA NIM** (`nvidia`) - OpenAI-compatible endpoint (free for development/prototyping for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise licensing)
+- **NVIDIA NIM** (`nvidia`) - OpenAI-compatible endpoint
 
 ### Anthropic API Format
 
diff --git a/main/llm/llm_proxy.c b/main/llm/llm_proxy.c
index b7189590..3181e34d 100644
--- a/main/llm/llm_proxy.c
+++ b/main/llm/llm_proxy.c
@@ -161,6 +161,7 @@ static const char *llm_api_url(void)
 /** Return the HTTP Host header value for the active provider. */
 static const char *llm_api_host(void)
 {
+    /* TODO(PR#40): derive host from configured API URL once custom provider URLs are supported. */
     if (provider_is_openai()) return "api.openai.com";
     if (provider_is_openrouter()) return "openrouter.ai";
     if (provider_is_nvidia()) return "integrate.api.nvidia.com";

From c882fee463f268e2d393200a5313d7da92db92a1 Mon Sep 17 00:00:00 2001
From: Mauro Druwel <mauro.druwel@gmail.com>
Date: Sun, 15 Mar 2026 11:08:52 +0100
Subject: [PATCH 5/5] coderabbit suggestions

---
 main/llm/llm_proxy.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/main/llm/llm_proxy.c b/main/llm/llm_proxy.c
index 2bd77db0..cd486d7b 100644
--- a/main/llm/llm_proxy.c
+++ b/main/llm/llm_proxy.c
@@ -293,7 +293,7 @@ static esp_err_t llm_http_direct(const char *post_data, resp_buf_t *rb, int *out
 
     esp_http_client_set_method(client, HTTP_METHOD_POST);
     esp_http_client_set_header(client, "Content-Type", "application/json");
-    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) {
+    if (provider_is_openai_compat()) {
         if (s_api_key[0]) {
             char auth[LLM_API_KEY_MAX_LEN + 16];
             snprintf(auth, sizeof(auth), "Bearer %s", s_api_key);
@@ -321,7 +321,7 @@ static esp_err_t llm_http_via_proxy(const char *post_data, resp_buf_t *rb, int *
     int body_len = strlen(post_data);
     char header[1024];
     int hlen = 0;
-    if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) {
+    if (provider_is_openai_compat()) {
         hlen = snprintf(header, sizeof(header),
             "POST %s HTTP/1.1\r\n"
             "Host: %s\r\n"