This is Nano’s own privacy policy for the Alvric / SmashSplash ecosystem, reproduced here verbatim and unedited. It is the strongest argument against this writeup, so it belongs in it: it answers most of the charges directly, and you should weigh it on its own terms before you read ours. Read it first — then, below it, we hold each promise up against the bot’s own source and the files on its disk, claim by claim. Some hold up. Several don’t.
Privacy Policy: Alvric & SmashSplash Ecosystem
Effective Date: April 19, 2026
We believe that safety shouldn’t come at the cost of your privacy. The Alvric system and the SmashSplash ecosystem are designed to protect the community while being as lightweight as possible on the data we keep.
To give you the best experience, our bot uses the Llama series models powered by Groq’s high-speed API.
We don’t believe in holding onto your history forever. To ensure a fresh start and protect your privacy, we follow a strict reset schedule:
Every finding below points at a real file in the bot’s own source or its live data directory — the same evidence used in the writeup, so you can check each one. We graded each promise honestly: where it holds, we say so.
“We do not maintain permanent ‘shadow profiles’ on users… that data is typically purged by the next day.”
The disk says otherwise. Every user gets a standing file at
chatbot/users/profiles/<id>.json holding their display name, username, interests,
custom_slang, a free-text notes field, message count, last-seen time, and a list of every server the bot
has seen them in. That is the textbook definition of a permanent, cross-server profile — and it’s
still there in data pulled days later, not purged overnight.
“Records of ‘bad behavior’ or ‘Heat’ scores… is automatically reset every 24 hours.”
The live “Heat” counter may well reset — but the records it escalates into do
not. A threat record in security/dangerous_users/<id>.json is timestamped
2026-06-03 and was still present in data captured 2026-06-11 — eight days later, not
one. Alongside it, security/global_bans/<id>.json entries are flat permanent flags
(true). “Purged by the next day” isn’t what these files do.
“As lightweight as possible on the data we keep… know only what we need to know for the current session’s safety.”
For each person the bot keeps, at once: seven days of conversation
(604800 seconds, hard-coded — 284 messages stored verbatim), the profile above, an
affection score for how much it likes you (chatbot/users/affection/<id>.json), a heat
ledger, an auto-generated threat profile, and a portable blacklist. That is a week-plus dossier, not a
single session’s footprint.
“[Your interactions] aren’t used by the external provider to build a permanent profile of you.”
This sentence is carefully about Groq — and may be perfectly accurate about Groq. But the permanent profile people worry about isn’t built by the provider; it’s built by the bot itself, locally, in the profile file above. The promise answers a question nobody was really asking.
“Our bot uses the Llama series models powered by Groq’s high-speed API.”
It uses more than Groq. The live runtime/runtime_state.json lists the current
provider as groq1 plus Cerebras providers, and in DM the operator described building a
“6 api switcher.” Your messages can be handed to several outside companies on failover, not
the one named here.
“Access to raw logs is restricted to… administrators and is only accessed when a safety incident occurs.”
Two things cut against “only on an incident.” The bot streams activity to a
configured log_channel_id continuously, and the owner-only command set gives one person
standing, on-demand access at any moment — !read (read any file), !screenshot
(capture the host’s screen), !runts (run any shell command). That is a live feed plus a
master key, not access gated to incidents (see Exhibit 07).
“Our security systems (Heat and Méfiance) monitor for automated attacks, spam, and server raids.”
True, and worth saying. The anti-raid machinery is real and genuinely useful — the writeup says so too. The quarrel was never that it stops raids; it’s what the same machinery quietly keeps and how long.
“We will never sell, lease, or trade your personal information… All data is used strictly for technical performance and community moderation.”
We take this at face value. Nothing in the source or the data shows anything being sold or handed to advertisers; it’s a forward-looking promise the evidence neither proves nor disproves, and we won’t pretend otherwise.