---

NanoGPT SillyTavern Setup: Complete 2026 Guide

SillyTavern is the go-to frontend for AI roleplay and creative writing. NanoGPT gives you access to 400+ models through a single API — no subscription, pay-per-prompt, crypto payments accepted.

This guide walks you through connecting NanoGPT to SillyTavern. From API key to first message.

What You Need Before Starting

Three things:

1. A NanoGPT account with credits loaded (minimum $1 to start)
2. SillyTavern installed on your machine (Windows, Mac, or Linux)
3. 5 minutes — that's how long the actual setup takes

If you don't have SillyTavern yet, grab it from the official GitHub: github.com/SillyTavern/SillyTavern. Follow their install guide, then come back here.

If you don't have NanoGPT, sign up here. You can deposit crypto (Monero, Bitcoin, Nano) or use a card. No KYC required.

Step 1: Get Your NanoGPT API Key

Log into your NanoGPT dashboard at nanogpt.com.

1. Click your profile icon (top right)
2. Select "API Keys"
3. Click "Generate New Key"
4. Copy the key — it starts with ng-

Important: Save this key somewhere safe. NanoGPT shows it once. If you lose it, you'll need to generate a new one.

Key Permissions

When generating the key, you'll see permission options:

• Chat Completions — Enable this (required for SillyTavern)
• Image Generation — Optional, only if you want AI images in SillyTavern
• Model Listing — Enable this (helps SillyTavern discover available models)

Leave the rest unchecked. Fewer permissions = less risk if the key leaks.

Step 2: Configure SillyTavern for NanoGPT

Open SillyTavern in your browser (usually http://localhost:8000).

1. Click the API Connection icon (plug icon, top bar)
2. Under "API", select Chat Completion (not Text Completion)
3. Under "Chat Completion Source", select Custom (OpenAI-compatible)
4. Fill in the fields:

Field	Value
Custom Endpoint URL	https://api.nanogpt.com/v1
API Key	Your ng-... key from Step 1
Model	Leave as default for now (we'll set this next)

5. Click Connect (or the checkmark button)

If the connection turns green, you're connected. If it's red, check the troubleshooting section below.

Model Selection for SillyTavern

Once connected, SillyTavern should show a dropdown of available models. If not, click "Refresh" next to the model field.

Best models for SillyTavern use:

For creative writing:

• claude-3.5-sonnet — Best prose quality, handles complex scenarios
• gpt-4o — Strong all-rounder, good with dialogue

For roleplay:

• llama-3-70b — Great character consistency, cheaper than Claude
• mixtral-8x22b — Good balance of quality and speed

For budget use:

• deepseek-v3 — Excellent value, surprisingly good for RP
• gemini-2.0-flash — Fast responses, good for quick interactions

Step 3: Test Your Connection

Create a new chat in SillyTavern (or use an existing character).

Type a simple message like: "Hello, introduce yourself."

If everything works:

• You'll see a response within 5 — 15 seconds
• The model name appears in the chat header
• No error messages in the SillyTavern console

If it doesn't work, check the troubleshooting section.

Best NanoGPT Models for SillyTavern

I've tested every major model on NanoGPT with SillyTavern. Here are my picks:

Creative Writing: Claude 3.5 Sonnet

• Quality: 9/10
• Speed: Medium (5 — 15 seconds per response)
• Cost: ~$0.003 — $0.01 per message
• Best for: Long-form narratives, complex character development, emotional depth

Claude writes like a human author. It handles nuance, subtext, and character voice better than any other model I've tested.

Roleplay: Llama 3 70B

• Quality: 8/10
• Speed: Fast (3 — 8 seconds per response)
• Cost: ~$0.001 — $0.003 per message
• Best for: Consistent character portrayal, action scenes, group chats

Llama 3 stays in character better than most proprietary models. It's also significantly cheaper.

Budget Pick: DeepSeek V3

• Quality: 7.5/10
• Speed: Fast (2 — 5 seconds per response)
• Cost: ~$0.0005 — $0.001 per message
• Best for: Casual chats, testing scenarios, high-volume use

At a fraction of a cent per message, DeepSeek is unbeatable for value. Quality is lower than Claude, but for everyday use it's more than good enough.

Speed Demon: Gemini 2.0 Flash

• Quality: 7/10
• Speed: Very fast (1 — 3 seconds per response)
• Cost: ~$0.0003 — $0.001 per message
• Best for: Quick back-and-forth, real-time conversations, testing setups

If response time matters more than prose quality, Flash is your model.

Troubleshooting Common Errors

"API Key Invalid"

Cause: Key is wrong, expired, or has insufficient permissions.

Fix:

1. Go back to NanoGPT dashboard → API Keys
2. Verify the key is active (not revoked)
3. Copy the full key again (no extra spaces)
4. Make sure "Chat Completions" permission is enabled
5. Paste into SillyTavern and reconnect

"Rate Limited"

Cause: Too many requests in a short time.

Fix:

1. Wait 30 seconds, then retry
2. In SillyTavern settings, increase the "Reply Cooldown" (under Generation Settings)
3. If you're on NanoGPT's pay-per-prompt plan, rate limits are generous ($8/month plan has even higher limits)

"Model Not Found"

Cause: Model name is misspelled or the model isn't available on NanoGPT.

Fix:

1. Click "Refresh Models" in SillyTavern's API settings
2. Select from the dropdown instead of typing manually
3. Check NanoGPT's model list at nanogpt.com/models

Slow Responses

Cause: You're using a large model or NanoGPT's servers are busy.

Fix:

1. Switch to a faster model (DeepSeek or Gemini Flash)
2. Reduce max_tokens in SillyTavern's generation settings (try 300 — 500 for shorter responses)
3. Check NanoGPT's status page for outages

SillyTavern Extensions Not Working

Some SillyTavern extensions rely on specific API features. Here's compatibility:

Extension	Works with NanoGPT?
Text-to-Speech	Yes (if TTS model available)
Image Generation	Yes (NanoGPT has image models)
Summarize	Yes
Vector Storage	Yes
Author's Note	Yes
Regex Scripts	Yes (client-side, works with any API)

NanoGPT vs OpenRouter for SillyTavern

Both work as OpenAI-compatible endpoints for SillyTavern. Here's the difference:

Feature	NanoGPT	OpenRouter
Pricing	$8/month flat OR pay-per-prompt	Pay-per-token only
Crypto payments	Yes (XMR, BTC, XNO)	No (credit card only)
KYC	None	Required for some features
Model count	400+	200+
Privacy	High (no logs claimed)	Medium (US-based, logs possible)
Speed	Fast	Varies by provider

For SillyTavern specifically, NanoGPT wins on privacy and pricing. OpenRouter has slightly better uptime consistency.

For a full comparison, see our NanoGPT vs OpenRouter breakdown.

SillyTavern Preset Recommendations for NanoGPT

Beyond model selection, your SillyTavern preset (generation settings) matters. Here are tested presets for NanoGPT models:

For Claude 3.5 Sonnet:

• Temperature: 0.8 — 1.0
• Top-P: 0.95
• Max tokens: 800 — 1200
• Frequency penalty: 0.0 (Claude handles repetition well)

For Llama 3 70B:

• Temperature: 0.7 — 0.9
• Top-P: 0.9
• Max tokens: 600 — 1000
• Frequency penalty: 0.1 (helps with Llama's occasional loops)

For DeepSeek V3:

• Temperature: 0.8 — 1.0
• Top-P: 0.95
• Max tokens: 500 — 800
• Frequency penalty: 0.15 (DeepSeek can get repetitive)

Start with these and adjust to taste. The biggest quality lever is temperature — lower for consistency, higher for creativity.

Privacy Considerations

When using NanoGPT with SillyTavern, your conversations go through NanoGPT's servers. If privacy matters:

• NanoGPT claims no-log policies
• Use crypto payments (Monero) to avoid linking your identity
• Use a VPN or Tor when connecting to the API
• For maximum privacy, run a local model with Ollama instead

For a full breakdown of private AI options, check our anonymous AI chat guide.

FAQ

Is NanoGPT compatible with all SillyTavern features?

Yes. NanoGPT uses an OpenAI-compatible API, which SillyTavern fully supports. Chat completions, streaming, function calling — all work. Some niche features (like specific logit bias parameters) may not be supported, but you won't notice in normal use.

How much does NanoGPT cost for SillyTavern?

Depends on your usage. Casual use (20 — 50 messages/day): $1 — $3/month. Heavy use (200+ messages/day): $5 — $15/month. The $8/month flat rate plan covers unlimited use of select models, which is the best deal for power users.

Can I use NanoGPT with SillyTavern extensions?

Yes. Most extensions work because they interact with the API through SillyTavern's abstraction layer, not directly with the provider. TTS, image generation, summarization, and vector storage all work fine.

Does NanoGPT log my SillyTavern conversations?

NanoGPT claims they don't store conversation logs. This is better than OpenAI (which logs everything) and comparable to other privacy-focused providers. If you're concerned, use a local model through other privacy AI tools.

---

Tested with SillyTavern 1.12.x and NanoGPT API v2. Last updated: July 2026.

For more NanoGPT guides, visit AI Privacy Tools.