NanoGPT + SillyTavern Setup: The Privacy-First AI Chat Combo

if you're sending personal stuff to an AI - roleplay, creative writing, whatever - you probably don't want that tied to your real identity. SillyTavern running locally with NanoGPT as the backend is the best setup i've found for this. no cloud UI, no account linked to your phone number, just your machine talking to an API.

tl;dr: SillyTavern + NanoGPT gives you private AI chat with 50+ models for $5-9/month. setup takes 15 minutes with Node.js and one API key. conversations stay local, no phone number required, crypto payments accepted. ChatGPT Plus can't do any of this.

Key Takeaways:

  • SillyTavern + NanoGPT costs $5-9/month for 50+ models vs $20/month for ChatGPT Plus with one model family
  • setup takes 15 minutes: clone SillyTavern, add NanoGPT API key, pick a model, start chatting
  • privacy stack: SillyTavern stores chats locally, NanoGPT doesn't train on data, crypto payments keep identity separate

i've been running this combo since march 2026. here's exactly how to set it up and what i learned along the way.

why SillyTavern + NanoGPT

SillyTavern is open-source. it runs on your machine, stores chats locally, and connects to any OpenAI-compatible API. NanoGPT is that API - 50+ models, one key, crypto payments. together they give you private AI chat with model flexibility that ChatGPT can't touch.

the privacy stack looks like this:

  • SillyTavern - runs locally, no data sent to third parties by default
  • NanoGPT - no conversation training, crypto payments, email optional
  • your machine - chats stay on your disk, not someone's cloud

compare that to ChatGPT where every conversation is stored on OpenAI's servers, linked to your phone number and credit card. for casual stuff that doesn't matter. for personal creative work? it matters.

what you need

before we start, make sure you have:

  • Node.js 18+ - check with node --version
  • Git - for cloning the repo
  • a NanoGPT account with an API key (get one here)
  • 8 GB RAM minimum (SillyTavern is a bit hungry)

if you don't have Node.js, grab it from nodejs.org. the LTS version works fine.

step 1: install SillyTavern

linux / mac

open a terminal and run:

git clone https://github.com/SillyTavern/SillyTavern.git
cd SillyTavern
./start.sh

that's it. the script handles dependencies on first run. takes maybe 2-3 minutes depending on your internet.

windows

  1. download the installer from sillytavern.ai
  2. extract the zip
  3. double-click start.bat

it opens a browser window at http://localhost:8000. if it doesn't, just open that URL manually.

first run

SillyTavern will ask you to set a username and password. this is local only - it's not creating an account anywhere. just pick something you'll remember.

step 2: connect NanoGPT

this is the part people get wrong. it's actually simple.

  1. open SillyTavern in your browser (http://localhost:8000)
  2. click the gear icon in the top right (settings)
  3. go to API Settings (sometimes called "Connection Profile")
  4. set API Type to "OpenAI" or "ChatCompletion"
  5. in API Base URL, enter: 
    https://api.nano-gpt.com/v1
  6. in API Key, paste your NanoGPT API key
  7. click Connect (or "Check" / "Verify")

if it says "Connected" or shows a green indicator, you're done with the connection part. if it fails, triple-check the URL - no trailing slash, no typos. the most common mistake is typing nano-gpt.com instead of api.nano-gpt.com.

getting your API key

if you don't have one yet:

  1. go to nano-gpt.com and create an account
  2. top up at least $8 (minimum deposit, not a subscription)
  3. go to the API section in your dashboard
  4. generate a new key
  5. copy it immediately - you won't see it again

more detailed setup: NanoGPT API key guide

step 3: pick a model

after connecting, you need to select which model SillyTavern will use. in the model dropdown (usually near the API settings), you'll see a long list. here's what i recommend:

for text chat and roleplay

Claude 3.5 Sonnet - best quality for longer conversations. more nuanced, better at staying in character. costs about $3/1M input tokens. worth it.

GPT-4o - faster, good for shorter exchanges. slightly less creative but more consistent. $2.50/1M input.

for creative writing

Claude 3.5 Sonnet - no contest. it handles tone, pacing, and character voice better than anything else i've tried.

GPT-4o - stronger at technical accuracy.

for long conversations (100+ messages)

Gemini 1.5 Pro - the 1M token context window means it remembers what you said 200 messages ago. other models start losing context around 50-100 messages.

for budget usage

GPT-4o-mini - $0.15/1M tokens. good enough for casual chat, way cheaper. use this if you're chatting a lot and don't need top-tier quality.

these are the settings i use. tweak to taste.

for regular chat

SettingValue
Temperature0.7
Max Tokens500
Top P0.9
Frequency Penalty0.3

temperature 0.7 gives natural responses without being too random. max tokens 500 keeps responses concise - bump to 1000+ if you want longer answers.

for creative writing / roleplay

SettingValue
Temperature0.9
Max Tokens1000
Top P0.95
Frequency Penalty0.2

higher temperature = more creative but less predictable. 0.9 is the sweet spot for me. anything above 1.0 tends to get incoherent.

for code

SettingValue
Temperature0.3
Max Tokens2000
Top P0.9
Frequency Penalty0.0

low temperature for code. you want deterministic, correct output, not creative interpretations of your function.

creating characters

SillyTavern's real power is character cards. here's a quick rundown:

  1. click Character Management in the sidebar
  2. click Create New Character
  3. fill in:
  • Name - whatever you want
  • Description - personality, behavior, how they talk
  • First Message - the opening line
  • Scenario - context for the conversation

example character

Name: Luna
Description: Luna is a direct, no-nonsense assistant who gives 
short answers. She prefers bullet points over paragraphs. 
She occasionally makes dry jokes. She never apologizes for 
being blunt.
First Message: Hey. What do you need?
Scenario: Luna is a senior developer you can ping for quick 
advice. She has 15 years of experience and zero patience 
for vague questions.

the description matters more than people think. a well-written character description produces dramatically better responses than a vague one. spend 10 minutes on it.

the privacy angle

this is why i use this setup. let me be specific about what stays private:

what NanoGPT doesn't do

short answer: NanoGPT doesn't train on your conversations, require a phone number, or require a real email.

  • doesn't train on your conversations (stated policy)
  • doesn't require a phone number
  • doesn't require a real email (use a burner)
  • crypto payments mean no credit card trail

what SillyTavern doesn't do

short answer: SillyTavern doesn't send data anywhere except your configured API. no telemetry, no analytics, no phone-home.

  • doesn't send data anywhere except the API you configure
  • stores everything locally on your machine
  • no telemetry, no analytics, no phone-home

what you should still do

short answer: use a VPN to hide your IP, pay with Monero for true anonymity, use a burner email, and encrypt your SillyTavern folder.

  1. use a VPN - hides your IP from NanoGPT (or anyone else)
  2. pay with Monero - if you want true anonymity, XMR is the way. guide on buying Monero anonymously
  3. use a burner email for your NanoGPT account
  4. encrypt your SillyTavern folder - chats are stored as plain text files by default

is this overkill for most people? probably. but if you're reading this site, you probably care about this stuff. and the setup once done is just as easy as ChatGPT - you just click open SillyTavern and start chatting.

troubleshooting common issues

"connection failed"

  • check the base URL: https://api.nano-gpt.com/v1 (no trailing slash)
  • make sure your API key is valid and has balance
  • try regenerating the key if it still fails

responses are slow

  • switch to a faster model (GPT-4o-mini or Claude 3 Haiku)
  • check if your VPN is adding latency
  • NanoGPT doesn't queue during peak hours, so it's usually not their side

responses are garbage

  • try a different model. seriously. what looks like bad AI is often just a bad model-task match
  • adjust temperature (lower for accuracy, higher for creativity)
  • check your character description - a confusing system prompt produces confusing output

rate limited

  • wait 1-2 minutes
  • check your NanoGPT balance
  • reduce the frequency of requests

what this costs me

real numbers from my usage over 3 months:

  • march 2026 (testing, heavy use): $9.20
  • april 2026 (regular use): $5.40
  • may 2026 (regular use): $4.80

i chat maybe 30-40 messages a day, using Claude 3.5 Sonnet most of the time. compare that to $20/month for ChatGPT Plus where i'd have one model and no privacy.

the $8 minimum deposit lasts me about 2 weeks with regular use. credits don't expire. when i use less, the leftover carries over.

model switching in practice

the killer feature is switching models mid-session. here's what i actually do:

  • start a conversation with GPT-4o for quick back-and-forth
  • switch to Claude 3.5 Sonnet when the conversation gets interesting or creative
  • switch to Gemini 1.5 Pro when the context gets really long (100+ messages)
  • use GPT-4o-mini for throwaway questions

one SillyTavern instance, one NanoGPT key, four models. try doing that with ChatGPT.

final thoughts

SillyTavern + NanoGPT is the setup i recommend to anyone who cares about privacy and model flexibility. it's not as polished as ChatGPT's web UI - SillyTavern's interface has a learning curve and it's not winning design awards. but for actual functionality and privacy, nothing else comes close.

if you're already on BHW or Reddit and testing AI tools, you'll figure out SillyTavern in an afternoon. the community is active and there are character card repositories everywhere.

👉 Get a NanoGPT API key - $8 minimum, no subscription, crypto payments.

last updated: july 2026

Ready to swap crypto privately?

No KYC. No account. Instant swaps.

Swap Now