Configuration

ESP-Claw basic_demo stores boot defaults in menuconfig (sdkconfig) and runtime overrides in NVS.

basic_demo uses the NVS namespace basic_demo. On boot, menuconfig defaults load first, then keys present in NVS override them. Field list and read/write logic live in application/basic_demo/main/basic_demo_settings.c.

basic_demo settings

Unless marked optional, fields are required.

Basic

wifi_ssid / wifi_password: Wi‑Fi SSID and password
- Except ESP32-C5, only 2.4 GHz Wi‑Fi is supported.
- Empty password means open network.
time_timezone: time zone
- Must be a POSIX TZ string; see this table.
- Example for Beijing/Hong Kong/Singapore: CST-8.

LLM

The firmware LLM runtime (claw_llm_runtime.c inside claw_core) builds requests from profile (preset) + backend (protocol):

profile picks default gateway root, API path shape, max_tokens field names, etc.;
backend chooses OpenAI Chat Completions–compatible vs Anthropic Messages request/response parsing.

Fields:

llm_profile: vendor/scenario preset
- Empty string falls back to claw_llm_profile_default() (currently openai).
- Mapping of preset names to concrete settings is in basic_demo_settings.c:24.
llm_backend_type: backend type [optional]
- Must match the real wire protocol or be empty: use anthropic for official Anthropic or Messages-compatible gateways; use openai_compatible for OpenAI or “OpenAI-compatible” gateways.
- Non-empty overrides the profile’s default backend; empty keeps the profile default.
- Allowed values: openai_compatible, anthropic, custom¹.
llm_base_url: API base URL [optional]
- Custom gateway root; empty uses the profile’s default base_url.
- Runtime endpoint is base_url + api_path.
llm_model: model name
- Must match the provider’s model id.
llm_api_key: LLM API key
llm_auth_type: auth header style [optional]
- Extra hint such as bearer, api-key, none, …
- With openai_compatible, empty llm_auth_type implies bearer and Authorization: Bearer <API Key>.
- With anthropic, llm_auth_type is ignored; the key goes in x-api-key: <API Key>.
llm_timeout_ms: request timeout
- Timeout in milliseconds (stored as string).

IM and platforms

ESP-Claw can connect QQ, Feishu, Telegram, and WeChat ClawBot, with extension hooks. If credentials for a channel are empty, that IM cannot work end-to-end.

qq_app_id / qq_app_secret: QQ Open Platform bot credentials
feishu_app_id / feishu_app_secret: Feishu app credentials
tg_bot_token: Telegram bot token
wechat_token / wechat_base_url / wechat_cdn_base_url / wechat_account_id: WeChat ClawBot fields, usually filled by the QR login flow.

Web search

Search API keys are optional. If both are empty, ESP-Claw cannot fetch online results via search.

search_brave_key: Brave Search API key—see Brave auth docs.
search_tavily_key: Tavily key—see Tavily quickstart.

Compile-time Kconfig options

Besides the NVS settings above, some behavior is controlled at build time via Kconfig (menuconfig) with no runtime toggle—change requires rebuilding and reflashing:

BASIC_DEMO_MEMORY_MODE: long-term memory mode.
- Structured (full structured memory, default): enables structured memory extraction; the LLM-visible groups additionally include claw_memory; the summary-tag catalog is injected instead of the full MEMORY.md text.
- Lightweight: skips structured extraction and injects MEMORY.md text directly—useful when RAM or context window is tight.
BASIC_DEMO_ENABLE_EMOTE: whether to start the LCD emote/status UI (app_expression_emote) at boot. Only meaningful on boards with an LCD; disabling saves a little heap.

Use idf.py menuconfig → (Top) → Basic Demo Config to view or change these options.

Where to edit

Web: see Web configuration.
menuconfig: good for factory Wi‑Fi/keys before mass flash; still overridden by NVS.
Web flasher: values entered in the wizard are saved to NVS like Web saves.

After changes, rebooting the device is recommended.

Custom backend for advanced use; read the source for details. ↩