overtchat

Open source · self-hosted

Your AI chat.Actually yours.

OvertChat brings local and hosted models into one polished, self-hosted app. You decide where requests go; accounts, conversations, files, and memories stay on your server.

View on GitHubQuick start
No OvertChat cloud Multi-user Realtime local voice
Continue
Bring your modelsOllamavLLMllama.cppSGLangLM StudioHosted APIs

Built to be used

Your models deserve an actual front door.

An inference server is infrastructure. OvertChat is the part people use: you configure the models once, then everyone you trust gets an account and a private space of their own.

01

Bring any model

Connect vLLM, llama.cpp, SGLang, Ollama, LM Studio, a hosted provider, or any OpenAI-compatible endpoint once at the server.

02

Choose who gets in

Create accounts for people you trust. Public signup stays closed, and every person gets private history, projects, files, and memories.

03

Let them just chat

They sign in from the web or Android app. Nobody else needs an inference URL, a shared API key, or a lesson in your model stack.

Beyond the text box

Local voice, at conversation speed.

Speak naturally. Cut in mid-answer. Change direction and keep going. OvertChat wraps local Parakeet and Kokoro around the model you choose, with web search and a saved transcript along for the conversation.

<8 GBVRAM for the entire speech layer in our current NVIDIA testing. The rest belongs to the model.

A 24 GB card can still have room for a capable quantized model. Speech can also run on CPU or a second GPU when you want to go bigger.

A focused Open WebUI alternative

Everything around the model matters too.

OvertChat stays focused on a fast, complete chat experience instead of becoming another platform to maintain. The everyday details are already handled.

Leave. The reply won’t.

Generations keep running on the server when a tab closes or a phone backgrounds, then reconnect when you return.

Useful search

Search the web through bundled SearXNG and find old conversations with SQLite FTS5—no vector database required.

Files, not just prompts

Work with images, PDFs, Word and Excel documents, CSV files, and source code directly in a chat.

One server, personal to everyone

Private history, projects, saved memories, and preferences make a shared installation feel like each person’s own app.

Tools and reasoning included

Connect standard MCP servers, use built-in web search, and expose the reasoning controls supported by each local or hosted model.

A side door to coding agents

When useful, admins can run Codex, Claude Code, OpenCode, Pi, and Oh My Pi locally or over SSH without leaving the chat app.

What “yours” means

No mystery cloud in the middle.

Accounts, conversations, files, memories, and settings stay in one portable SQLite database on your server. Model requests go directly to the endpoints you configure. OvertChat operates no account system, API relay, or hosted control plane.

Read the deployment guide
  • One portable SQLite file
  • No hosted control plane
  • No usage analytics
  • Optional local speech and search
  • Standard MCP support
  • One-command managed updates

Wherever there’s a browser

Open it. Pick up where you left off.

Long conversations stay responsive, active replies survive a disconnect, and every project, file, search, memory, and model is waiting on the server.

Self-host the web app

Away from the desk

Same server. Smaller screen.

The native Android app connects straight to OvertChat. Chats, attachments, projects, search, and voice follow without moving into a separate mobile backend.

Get it on Google Play

Your turn

One command between here and yours.

The guided setup handles Docker, secrets, updates, and optional search, speech, and Agent Connections. Bring a model endpoint; the first account becomes the administrator.

Full deployment guide
terminal
curl -fsSL https://overtchat.com/install | sh

Still becoming yours

Built in public. Shipped in the open.

Follow every stable web and mobile release in one chronological log, generated directly from the project’s GitHub Releases.

Browse releases