nanochat

Open-source self-hostable chat client for Nano-GPT.

Test it out at nanochat.app

Get 25 Free Daily prompts using any nano-gpt subscription model without needing an API key

Native Apps

Get the native NanoChat experience on your devices:

• Android: nanochat-android
• iOS: nanochat-ios (TestFlight Beta)
• Desktop: nanochat-desktop - Linux Only for now

---

Changes with this fork

• Convex -> SQLite + Drizzle
• Docker + Docker Compose
• Yarn -> Bun
• Openrouter -> Nano-GPT (nano-gpt.com)
• Theme inspired by T3 Chat
• Nano-GPT Web Search / Deep Search (Linkup / Tavily / Exa / Kagi)
• Nano-GPT Web Scraping when you enter a URL (adds to context)
• Nano-GPT Context Memory (Single Chat)
• Cross-Conversation Memory (All Chats)
• Nano-GPT Image Generation + img2img support
• Nano-GPT Speech-to-Text (Whisper/Wizper/ElevenLabs)
• Passkey support (requires HTTPS)
• Nano-GPT Video Generation
• Selectable System Prompts (Assistants)
• KaraKeep Integration (Thanks to jcrabapple)
• Nano-GPT YouTube Transcripts (Thanks to thejudge22)
• Follow-up Questions - Contextual follow-up question suggestions generated by LLM after each response.
• Configurable System Themes
• Model Performance Tracking and Analytics.
• Projects
• Benchmark Data from artificialanalysis.ai API
• Provider Selection for Models (NanoGPT)
• Nano-GPT Video Generation

Setup (Docker)

Installation

• Clone the repository git clone https://github.com/nanogpt-community/nanochat.git
• cd nanochat
• cp .env.example .env
• Edit the .env file with your configuration
• docker compose up

Setup (Bun)

Installation

• Install Bun (https://bun.sh/)
• Clone the repository git clone https://github.com/nanogpt-community/nanochat.git
• cd nanochat
• cp .env.example .env
• Edit the .env file with your configuration
• bun install
• bun run dev
• run npx drizzle-kit push to upgrade your database schema when new features are added!

Nginx

Ensure to have the following in your server block if you use nginx:

proxy_buffer_size 256k;
proxy_buffers 4 256k;
proxy_busy_buffers_size 256k;
client_max_body_size 50M;

---

Features Overview

Follow-up Questions

The follow-up questions feature automatically generates 2-3 contextual questions after each AI response. Key details:

• Generation: Uses zai-org/GLM-4.5-Air model via Nano-GPT
• Display: Shows 1 second after message generation completes
• Persistence: Suggestions are stored in the database and shown when loading historical conversations
• User Control: Can be toggled on/off in Account Settings
• Length Check: Only generates for assistant messages over 100 characters
• Interaction: Clicking a suggestion inserts it into the input field
• Cleanup: Suggestions are hidden when user sends a new message

Web Search

Multiple search providers available:

• Linkup - Standard and deep web search
• Tavily - Optimized for AI applications
• Exa - Neural search engine
• Kagi - Premium search results

YouTube Transcripts

Automatically fetch and transcribe YouTube videos when URLs are detected in user messages. Costs $0.01 per transcript.

Memory Systems

• Context Memory: Compresses long conversations within a single chat for better context retention
• Persistent Memory: Remembers facts about the user across different conversations

Assistants

Create custom system prompts with:

• Custom name and instructions
• Default model selection
• Default web search mode
• Web search provider selection

Image Generation

Generate images using Nano-GPT's image models with support for:

• Text-to-image
• Image-to-image (img2img)

Text-to-Speech

Listen to assistant messages read aloud using a variety of models:

• Models: OpenAI (TTS-1, HD), Kokoro (Multilingual), ElevenLabs (Premium)
• Controls: Play/Stop, Speed Control (0.25x - 4.0x)
• Cost Efficient: Supports ultra-low cost models like GPT-4o Mini TTS ($0.0006/1k)

Speech-to-Text

Transcribe voice messages using:

• Models: Whisper Large V3 (OpenAI), Wizper (Fast), ElevenLabs
• Usage: Click the microphone icon in the chat input
• Analytics: Usage and costs are tracked in Model Analytics

KaraKeep Integration

Save conversations as bookmarks to your KaraKeep instance for long-term storage and organization.

Model Benchmarks

View performance benchmarks from Artificial Analysis directly in the model picker:

• For LLMs: Intelligence Index, Coding Index, Math Index, and Speed (tokens/sec)
• For Image Models: ELO rating and Rank
• Benchmarks appear in the model info panel (click the info icon on any model)
• Requires ARTIFICIAL_ANALYSIS_API_KEY environment variable

Provider Selection

For models supported by multiple providers on NanoGPT, you can:

• Select a specific provider (e.g., 'openai', 'anthropic', 'google') for a model.
• Configure preferred and excluded providers in Account Settings.
• Enable automatic fallback to other providers if the preferred one fails.

Video Generation

Generate videos using NanoGPT's video models:

• Text-to-video generation
• View generation status and history
• Download generated videos

URL Parameters & Shortcuts

You can use URL parameters to pre-configure your chat session. This is useful for creating bookmarks or "bang" style shortcuts (e.g. in your browser).

Parameter	Description	Example
q	Pre-fills the chat input	?q=Explain quantum physics
model	Selects the AI model	?model=zai-org/glm-4.7
model_provider	Selects the provider for the model, or clears to auto with auto	?model_provider=cerebras
search	Sets web search mode (off, standard, deep)	?search=deep
search_provider	Sets search provider (linkup, tavily, exa, kagi, perplexity, valyu, Brave modes)	?search_provider=brave-research
search_context_size	Sets shared search context size (low, medium, high)	?search_context_size=high
search_exa_depth	Sets Exa depth (fast, auto, neural, deep)	?search_exa_depth=neural
search_kagi_source	Sets Kagi source (web, news, search)	?search_kagi_source=news
search_valyu_search_type	Sets Valyu search type (all, web)	?search_valyu_search_type=web
projectId	Contextualizes chat with a specific Project	?projectId=123

Example "Bang" URL: https://nanochat.app/chat?model=zai-org/glm-5.1&search=deep&q=%s

In-Chat Shortcuts:

• @<rule_name>: Apply a specific user rule to the current message (e.g., @concise)

---

Environment Variables

Variable	Description
DATABASE_URL	SQLite database path (default: ./data/nanochat.db)
NANOGPT_API_KEY	Nano-GPT API key for generation
BETTER_AUTH_SECRET	Authentication secret
BETTER_AUTH_URL	Base URL for authentication
ARTIFICIAL_ANALYSIS_API_KEY	(Optional) API key for model benchmarks from artificialanalysis.ai
API_KEY_HASH_SECRET	(Optional) Dedicated secret for developer API key lookup hashes; defaults to ENCRYPTION_KEY or BETTER_AUTH_SECRET
ENCRYPTION_KEY	Encryption key for API keys and other stored secrets at rest. Generate with openssl rand -base64 32

API Key Encryption

The application supports encrypting API keys stored in the database using AES-256-GCM:

• Required for new secrets: ENCRYPTION_KEY must be set before creating developer API keys, provider keys, or other stored secrets
• At rest: Secrets are encrypted with AES-256-GCM
• Developer API auth: API keys are also indexed with a non-reversible lookup hash so requests no longer require decrypting the entire key table
• Schema update: Run npx drizzle-kit push after upgrading so the api_keys.key_hash column exists
• Migration: Run bun run scripts/migrate-encrypt-api-keys.ts to encrypt existing keys
• Details: See scripts/README-API-KEY-ENCRYPTION.md

---

Database Schema

The application uses SQLite with Drizzle ORM. Key tables:

• messages: Stores chat messages with content, role, annotations, and follow-up suggestions
• user_settings: User preferences including follow-up questions toggle
• conversations: Chat sessions with metadata
• assistants: Custom system prompts
• user_memories: Persistent cross-conversation memory

---

Tech Stack

• Frontend: SvelteKit + Svelte 5
• Styling: Tailwind CSS
• Database: SQLite + Drizzle ORM
• Auth: Better Auth
• AI Provider: Nano-GPT (nano-gpt.com)
• Runtime: Bun
• Container: Docker

---

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

MIT License - See LICENSE file for details.