Pulpo

Open-source chatbot for everyday use

Pulpo comes with native mobile apps, fast disposable workspaces, and broad model support. Everything stays perfectly in sync.

DownloadOpen in your browser

Use any model

Run open-weight models on your own GPUs, so conversations can stay on your network, or use the hosted APIs you already pay for. Chat, image, and speech models share the same provider connections, and you can switch between them in the middle of a conversation.

Models

  • OpenAI
  • Claude
  • Gemini
  • Grok
  • Mistral
  • DeepSeek
  • Qwen
  • Kimi
  • GLM

Runs on

  • Your own hardware
    • vLLM
    • llama.cpp
    • Ollama
  • Azure AI Foundry
  • Fireworks
  • Baseten
  • OpenRouter
  • Any compatible endpoint
Runs on your own hardware
Pulpo speaks OpenAI Responses, Chat Completions, Anthropic Messages, and Mistral, so any inference server that offers one of them works, from vLLM and SGLang to llama.cpp and Ollama. Chats, agents, and the API behave the same on each.
Inference clouds, too
Connect Azure AI Foundry, including its image models, Fireworks with prompt-cache affinity, Baseten, and OpenRouter alongside each lab's own API.
Controls for each model
Reasoning effort, speed, and other presets sit in the composer, and Pulpo remembers your choice for every model.

A fresh computer for every task

Turn on Agent mode and the model gets its own Linux machine to run code, crunch data, and build files. Workspaces are claimed from a warm pool, so they start in an instant.

Isolated
Every session runs in its own lightweight virtual machine, with nothing shared between chats or people.
Ready to work
Python, Node.js, Git, and FFmpeg come preinstalled, along with tools for spreadsheets, PDFs, Word, and PowerPoint.
Disposable
The agent has root, and nothing it does can outlive the session. What you want to keep goes to your Files.

Native apps for iPhone and Android

Built with SwiftUI and Liquid Glass on iOS and Material 3 Expressive on Android.

And everything else you'd expect

The everyday parts of a chat app are all here, on every platform.

  • Files and documents

    Folders, uploads, and collaborative documents live next to your chats, and agents can work on them.

  • Memory and search

    Pulpo can recall earlier conversations and search them by meaning, not just by keyword.

  • Voice

    Dictate a message, or have any answer read aloud with playback controls.

  • Image generation

    Agents create images with the image models you connect.

  • Quizzes and flashcards

    Ask to be quizzed and the reply becomes an interactive quiz with hints and a score.

  • Sharing

    Publish a conversation as a public link that anyone can read.

  • Developer API

    OpenAI- and Anthropic-compatible endpoints, with keys you issue yourself.

  • Passkeys and two-factor

    Sign in with a passkey or protect your account with a second factor.

Open source, and yours to run

The whole stack is on GitHub: server, web, desktop, mobile, and the workspace infrastructure. Read it, change it, and run it on your own hardware.

One Docker Compose file
Postgres, Redis, object storage, the API, the worker, and the web app start together.
Run it your way
Manage providers, models, people, and budgets from the admin panel or the pulpo command line.
Your data stays yours
Provider keys are encrypted at rest, and backups can go offsite encrypted with a key only you hold.
View on GitHubRead the docs

Free for personal and noncommercial use under the Pulpo Noncommercial License.

Self-host in four commands
$ git clone https://github.com/IsaacThoman/pulpo.git
$ cd pulpo
$ cp .env.example .env
$ docker compose up --build -d
# http://localhost:8080

Start using Pulpo

Download the app for your device, or open it in the browser you already have.

DownloadOpen in your browser