Optional features

The example config has one Discord character. Everything below is added to jingmei.config.json as needed; each snippet says which environment variable it needs (put it in .env, key: value format). Field meanings are in the Configuration reference.

Telegram: needs TELEGRAM_LUNA_TOKEN in .env (the BotFather token). Add a top-level telegram section and a telegram account on the persona; setup steps are in Telegram setup.

json

"telegram": { "chatIds": ["-1001234567890"] }

json

"telegram": { "tokenEnv": "TELEGRAM_LUNA_TOKEN" }

The second snippet goes in the persona object under personas[], next to discord.

Voice (Fish Audio): needs FISH_AUDIO_API_KEY. referenceId is a 32-hex voice ID; a persona uses voice by default and can opt out with voiceEnabled: false.

json

"voice": {
	"apiKeyEnv": "FISH_AUDIO_API_KEY",
	"referenceId": "00000000000000000000000000000000",
	"model": "s2.1-pro-free"
}

Drawing: needs no environment variable, but an Antigravity OAuth login; steps in Drawing. The model defaults to gemini-3.1-flash-image, so write this only to change it:

json

"imageGeneration": { "model": "gemini-3.1-flash-image" }

Vision model: describes images with another model when a persona's main model cannot see images. Needs credentials for that provider (bun run jingmei login <provider> or its API key variable).

json

"visionModel": "openrouter/google/gemini-2.5-flash"

Candlestick charts and long text as an image: no environment variable; both are off by default. See Candlestick charts and Long text as an image.

json

"kline": { "enabled": true },
"textImage": { "enabled": true, "thresholdChars": 300 }

Feature switches (features): everything is on by default. Each of these can be turned off alone, which removes the tools and prompt lines and stops loading dependencies nothing uses.

json

"features": { "history": false, "memory": false, "soul": false, "search": false, "audit": false }

Switch	When `false`
`history`	No message index, no `related_messages` / `search_history`, no "（相关 N 条）" on context lines; the ~96 MB embedding model is not downloaded and fastembed and sqlite-vec are not loaded. Combining it with `events` is a configuration error
`memory`	No `remember_member_fact` / `recall_member_memory`, and no passive collection of member profiles, preferences, relationships or birthdays stated in chat. Kept: `/memory` and `/forget` still show and delete existing data, and `/birthday` still records a birthday by hand (which `celebrations` birthday greetings rely on)
`soul`	No `update_soul`, and stored private soul notes are no longer injected into the system prompt; stored data stays in the database
`search`	No `search_web` and no prefetched search results; `DEEPSEEK_API_KEY` remains the default model's credential
`audit`	No natural-tone audit of final text (the leak-marker check still runs); reply decisions are controlled separately by `jev.replyDecision`

run_js is always on. Configuration edits trigger a graceful reload; new switches apply after current conversations finish.

Jev (quick reactions, memory ranking; participation decisions use it too): remote Jev needs TYPESAFE_API_KEY; omit apiKeyEnv to use only the in-process wrapper, which then requires DEEPSEEK_API_KEY (or a localJev section), otherwise startup fails with a configuration error. See Jev.

json

"jev": {
	"apiKeyEnv": "TYPESAFE_API_KEY",
	"quickReactions": true,
	"memoryScoring": true
}

Topics (events): needs a resolvable decision client, that is the Jev section above or DEEPSEEK_API_KEY; the summaryModel provider needs credentials. The first start downloads an embedding model.

json

"events": { "summaryModel": "deepseek/deepseek-flash" }

Holiday and birthday greetings: no environment variable. space and channelId must be a server/channel you configured above; personaId is the character that posts.

json

"celebrations": [
	{
		"space": "discord:000000000000000000",
		"channelId": "000000000000000001",
		"personaId": "luna",
		"timeZone": "Australia/Sydney",
		"calendar": "both"
	}
]

Optional features ​

Optional features