Install ABot

This guide installs ABot as a local application using the published npm package. You do not need to clone the source repository or build the Web UI. ABot provides the runtime and provider adapters; you supply a model connection.

Before you start

You need Node.js 20.19.0 or later in the Node 20 line, or 22.12.0 and later, plus npm. Node 21 is not supported. Check your installed versions:

node --version
npm --version

Choose one model provider to begin with:

• Ollama: a running Ollama service with your chosen model already downloaded.
• OpenAI: an API key and a concrete model ID available to your account.

The model does not have to run on the same machine, but its endpoint must be reachable from the Runtime process. A browser that can reach an address does not prove that a container or WSL process can reach it.

Create your workspace

Run these commands in a terminal. The new my-abot directory will own your installation, configuration, and runtime state.

mkdir my-abot
cd my-abot
npm init -y
npm install @abot-ai/runtime
npx abot start

Keep that terminal running and open http://127.0.0.1:5177. ABot binds to your computer’s loopback interface by default. Do not expose that address to the public internet as part of this setup.

On a fresh installation, the Home setup wizard appears. Follow the complete setup walkthrough for every screen, field, and checkpoint. If you already have a configured installation, keep using its directory rather than creating another one each time you start ABot.

Connect your model

In the wizard, select OpenAI or Ollama and enter the exact model ID you intend to use. Set Context window (tokens) to a capacity the concrete model supports; the 32,768-token starter value is not automatic model detection. Avoid copying an example model name without checking that it is available from your provider.

For OpenAI, enter the API key in the setup form. ABot saves it in the project’s private .env, not in runtime JSON or in the response returned to the browser. Saving the connection itself does not make a paid model request; actually using a hosted model may incur provider charges.

For Ollama, install and start Ollama separately using its platform instructions. After choosing a model from its library, download it and check the server:

ollama pull <model-id>
curl http://127.0.0.1:11434/api/tags

Replace <model-id> with the real ID; do not include the angle brackets. The API response should list that model. These commands assume Ollama runs on the same host. If it does not, use an address reachable from ABot rather than assuming 127.0.0.1 means your desktop computer.

Finish the optional steps

The Embedding step configures optional semantic memory. You can skip it and return to Memory → Setup later. Test and save performs a small real request to the selected provider. A successful onboarding save also disables the separate Memory plugin, while retaining its stored data; skipped or failed embedding setup leaves that plugin selection unchanged.

Review Plugins carefully: fresh setup starts with all tools selected. Disable plugins or individual tools you do not need. Tool availability and permission to perform a sensitive action are separate controls.

The Computer step is optional and depends on where the Runtime runs. Desktop tools can require a Companion connection and operating-system permissions. Follow the platform-specific computer access instructions; you do not need to prepare desktop access merely to have a model conversation. Connecting a computer does not enable SPARK collection.

Choose Finish setup, then Start chatting. Your next step is a first conversation.

Start again later

Return to the same directory and run:

cd my-abot
npx abot start

Use the actual path if my-abot is not in your current directory. Keep your configuration, .env, and runtime state private. “Local Runtime” does not necessarily mean all processing stays local: hosted generation, remote embeddings, SPARK models, and external tools can send relevant data to their configured services.