Local & Self-Hosted Endpoints

It is very much possible to run a TomoriBot instance with everything local, from text generation to MCP utilities. Every endpoint below is OpenAI-compatible or a small reference wrapper, registered through the /providers commands and stored encrypted in the database (no .env needed). Pick a capability to set up:

Text & Embeddings

Local LLMAny OpenAI-compatible server such as Ollama, KoboldCPP, LM Studio, vLLM, or LiteLLM for text generation and embeddings.

Codex CLI via ChatMockA local OpenAI-compatible bridge that lets TomoriBot use Codex CLI models.

Image & Video

ComfyUIGenerate images and short videos through your own ComfyUI instance using an API-format workflow.

Voice

Text-to-SpeechLocal TTS engines (Chatterbox, Qwen3-TTS, IrodoriTTS, Fish S2 Pro, VoxCPM2, CosyVoice 3) for native Discord voice messages, voice cloning, and VoiceDesign.

Speech-to-TextTranscribe audio attachments locally with WhisperX, whisper.cpp, or KoboldCPP.

Web Search & Fetching

SearXNGPrivacy-respecting metasearch proxy as a local web_search tool to avoid public rate limits.

Crawl4AIBrowser-rendered markdown extraction for JS-heavy sites, powering the fetch_url tool.

Custom Tools

Local MCP ServerRun a local MCP server to extend TomoriBot with your own tools (self-hosted instances only).

Starting Local Servers with bun run launch

Instead of starting local servers manually, use bun run launch which starts the requested servers, waits for them to be ready, then launches the bot in watch mode automatically:

# Bot only, identical to bun run dev
bun run launch

# With the SearXNG and Crawl4AI Docker containers
bun run launch --searxng --crawl4ai

# With a local TTS server (venv must be set up first — see the Text-to-Speech guide below)
bun run launch --qwen3tts
bun run launch --fishs2
bun run launch --voxcpm2
bun run launch --cosyvoice3

# See all available flags
bun run launch --help

Available flags: --searxng, --crawl4ai, --qwen3tts, --chatterbox, --irodoritts, --voxcpm2, --fishs2, --cosyvoice3, --moss, --whisperx

Docker-backed servers (--searxng, --crawl4ai) are created on first run and reused on subsequent runs. Python TTS/STT servers require their venv to be set up once beforehand; see the individual setup guides under Text-to-Speech and Speech-to-Text.

Ctrl+C stops the bot and any Python server processes. Docker containers are intentionally left running. Stop them manually with docker stop searxng / docker stop crawl4ai when done.