Wire up AI that stays on.

Install

Three steps, then press On

1. 1

   Get the code

   Clone the repo, or download the ZIP and unpack it anywhere.

   git clone https://github.com/Boltjar/Boltjar.git

2. 2

   Start it

   On Windows, double-click start.bat. On macOS or Linux, run the start script from the folder.

   ./start.sh

3. 3

   Open the link it prints

   The editor runs at this address, and your browser opens it on its own when it can.

   http://127.0.0.1:8770

The first start needs Node.js 18 or newer, because it builds the editor once. No Python 3.11 to 3.13 on the machine? The start script fetches a portable Python 3.12 and checks its SHA-256 before it uses it.

Coming soonA one-line installer.

Install guide, options and manual setup →

What you can build

Six graphs to start from

Each one is a guide in the docs, wired from the nodes that ship with Boltjar.

• Chat assistant

  1. Chat Input
  2. Template
  3. LLM
  4. Chat

  Talk to a local model through Ollama, or to a cloud model with a key. The chat example keeps its history in a SQLite table.

  Runs locally with Ollama

• Webhook to LLM

  1. Webhook
  2. Template
  3. LLM
  4. Append File

  An endpoint on your machine. A POST fires the graph, the model sums up the body, and the summary goes wherever you wire it: a file, a chat service, a database.

  Runs locally with Ollama

• Scheduled digest

  1. Schedule
  2. HTTP Request
  3. LLM
  4. Write File

  Cron in a node. At 9 every morning, fetch a feed or an API, have the model sum it up, and save or send the result.

  Runs locally with Ollama

• RAG over your notes

  1. Chunk
  2. Embed
  3. Vectors
  4. LLM

  Chunk text, embed it with bge-m3 on Ollama, keep it in a vector store on disk, and hand the closest chunks to the model.

  Runs locally with Ollama

• Tool-calling agent

  1. LLM
  2. Tool
  3. Tool Args
  4. HTTP Request

  Tools are nodes. A Tool tells the model what it can call, the nodes behind it do the work, and the result goes back to the model.

  Runs locally with Ollama

• Voice

  1. Chat Input
  2. LLM
  3. Strip
  4. TTS

  Replies you can hear: the model writes, TTS speaks with an xAI, Fish Audio or ElevenLabs voice, and STT turns a clip back into text.

  Needs a provider key

How it works

ComfyUI's typed graph, with n8n's triggers

Every trigger, model, store, transform, condition and output is a node, and behaviour is how you wire them.

• A graph is a server

  Press On and it keeps running, triggers listening, until you turn it Off or stop Boltjar. Closing the browser tab does not stop it.

• Triggers push, data is pulled

  A trigger fires an event. The node it reaches pulls its inputs at that moment, so a Template never runs on a clock: it runs when the LLM needs its text.

• Typed pipes

  Every port has a type, and a wire only lands where the types fit. The editor checks it while you drag, and the server checks it again before the graph turns On.

• Nodes are Python

  A node is a class with a decorator. Drop a pack folder into packs/, restart, and its nodes sit next to the built-in ones.

Local-first

Your graphs run on your machine

• Runs on your computer

  The server listens on 127.0.0.1. Its API checks the Host, the Origin and a token your install creates, which keeps web pages you visit away from it.

• Your files stay yours

  Graphs, snapshots, stores and the keys you add in the editor live in the user/ folder, which git ignores.

• No telemetry, no account

  The editor talks only to your own Boltjar server and loads nothing from anywhere else. The server calls only what you set up: the models and URLs your wired nodes use, Ollama, and the providers and endpoints you connect, which it also asks for their model lists.

• Offline with Ollama

  Chat, tool calls and embeddings run on local models through Ollama. With no model picked, or a cloud model without its key, the LLM node answers with a mock reply, so a chat graph still runs.

How the server keeps other pages and machines out →

Support

Support Boltjar

Boltjar is free, and I build it in the open. If it saves you time or keeps something you care about running, sponsoring it on GitHub is the most direct way to keep it moving. Clear bug reports, node packs and doc fixes help just as much.

Sponsor on GitHubReport a bugWrite a node pack