Documentation
Boltjar is a visual node editor for systems that stay on. You wire nodes on a canvas: triggers such as a chat message, a webhook or a schedule; data such as text, templates and JSON; models for chat, speech and embeddings; stores for SQLite, key-value pairs and vectors; and outputs. Then you press On, and the graph runs as a server on your machine until you turn it Off.
Start here
Section titled “Start here”InstallGet the code, run the start script, open the link it prints.
Your first graphA chat with a model, in four nodes.
On, Off and RestartWhat turning a graph On checks, and what keeps running.
Triggers and pulled dataThe one idea that makes every graph readable.
Build something
Section titled “Build something”Each guide comes with the finished graph to download.
Chat with OllamaA local chat assistant, no key needed.
Webhook to LLMA POST in, a summary saved to a file.
Scheduled digestEvery morning at 9, a page summed up.
RAG with VectorsAnswers from your own notes.
Tool-calling agentA model that calls nodes as tools.
VoiceReplies you can hear. Needs a key.
Reference
Section titled “Reference”- Node reference: every node that ships, with its ports, types and knobs.
- Models: the model list your providers give, read live, why no model is ever picked for you, and how a manifest adds a model.
- Build a node pack: your own nodes, in Python, next to the built-in ones.
- Security: what the server lets in, and why a graph is code.