The gap
Cloud assistants send your questions to someone else's servers and lose your context between sessions. A plain local chat window keeps your data private, but it knows nothing about your files or projects.
AI
PN Scripts Assistant is a free, open-source assistant that learns your projects and documents and answers from that memory. It needs no database server, no containers and no cloud account, and what it learns stays on your disk.
Source code on GitHub under the MIT license. PN Scripts can adapt it for your project.
Where does that project live?
A real path on this disk, recalled from memory. Not a guess.
Every action is logged. A changed file can be restored.
The gap
Cloud assistants send your questions to someone else's servers and lose your context between sessions. A plain local chat window keeps your data private, but it knows nothing about your files or projects.
What it is
One Go program and one SQLite file. It uses a model you run in Ollama, learns from your code and documents with your approval, and answers with real file paths. What it learns never leaves your computer.
How you use it
Install it, point it at Ollama and ask. The first run finds Ollama, offers to install a model and asks where to keep the memory.
Ollama runs the language model. PN Scripts Assistant adds the long-term memory, the tools and the interface, and stores everything it learns in one file on your disk.
Use it when
Built for
What PN Scripts Assistant has learned about you never leaves your computer, in any mode and with any setting. There is no telemetry. If the privacy setting holds an unknown value, for example a typo, it is treated as private.
private Default
Nothing. Local model, local embeddings, no web access.
research
Search queries only. Conversations and memory stay here.
open
What you type may go to a third-party model. Memory stays here.
It recalls what it already knows about your projects and documents. Ask "Where does that project live?" and you get the real path on your disk.
It scans your code and documents and, after a conversation, suggests what is worth remembering. Anything it inferred waits for your approval. Facts it can check against the disk are saved automatically.
It can read files, list folders, run commands, search the web and control a smart home through Home Assistant. Every action is recorded, and a file it changed can be restored. In the Permissions panel you choose whether it asks every time, asks only about actions you have not allowed yet, or acts freely.
The interface draws a live map of what it has stored, grouped by how closely related the items are. You see what it knows at a glance.
In conversation mode it listens until you stop talking, answers, reads the answer aloud and listens again. Speech recognition and speech output both run locally. Voice can ask questions but cannot take actions, so a misheard sentence cannot change anything.
Without a graphics card, replies are slow. These figures are measured, so you know what to expect. The length of the prompt accounts for almost all of the reply time.
Prompt
~10 tok/s
Four cores, 7B model
Reply
~2.5 tok/s
Generation on CPU
Spoken turn
~40 s
Listen, answer, speak
Model loaded
6.8 s
15.8 s if the model reloads
Only a GPU, or open mode with a hosted model, changes the order of magnitude. Even in open mode, what PN Scripts Assistant has learned about you stays on your computer. Set OLLAMA_MAX_LOADED_MODELS to 2, or Ollama reloads the chat model before every reply.
Download a release from GitHub or build it from source with Go. You also need Ollama with a chat model and nomic-embed-text. The first run checks what is missing and offers to install it. Free and open source under the MIT license, on GitHub. PN Scripts can adapt it for your project.
Source code on GitHub under the MIT license. PN Scripts can adapt it for your project.