Pulp Free Chat User Documentation

paul · July 28, 2026, 3:48am

Pulp Free Chat — User Guide

Welcome to Pulp Free Chat, a private AI chat app that runs entirely on your own computer and home or office network. There are no accounts, no email, no sign‑up, and no cloud — your conversations stay on your devices.

This guide covers everything you need to get started and make the most of the app.

Current version: 1.1.24. Updated for changes since 1.1.17 — see What’s new since 1.1.17 for a summary.

What Pulp Free Chat is

Pulp Free Chat is a privacy‑first, local‑first AI assistant. All of the “thinking” (the AI inference) happens on your own hardware — nothing about your questions is sent to a company server.

It works in two parts:

The service runs on one computer (for example, your Mac, your Windows PC, or your Daisy AI computer). This is where the AI models actually run.
Thin clients — the app on your other devices (a laptop, phone, or tablet) — connect to that service over your local Wi‑Fi and show you the chat screen.

If you only use the app on the same computer where the service runs, you don’t need to think about any of this — it just opens and works.

Getting started

On the same computer

When you open Pulp Free Chat on the computer where the service is installed, it starts the service for you automatically. You’ll briefly see “Starting Pulp on this computer…”, then the chat screen loads. Press Continue and you’re in — no PIN or password needed on your own machine.

On another device (phone, tablet, or a second computer)

You can use Pulp Free Chat from another device as long as it’s on the same Wi‑Fi network as the computer running the service.

Step 1 — Show the connection code (on the host computer).
In the chat header, click the QR‑code icon (its tooltip reads “Connect a phone or tablet on your network”). A Connect a device window opens showing:

A QR code to scan.
The App address (something like https://your-computer.local:8443), with a copy button.
A Pairing PIN — a 6‑digit code you enter once per device.

Step 2 — Scan it (on your other device).
Open Pulp Free Chat on the phone or tablet, tap Scan QR code, and point the camera at the code on your computer. The app connects and trusts the local security certificate automatically — there’s no manual certificate step in the normal flow.

Prefer to type it in? On the connect screen, choose Enter address manually, type the App address (e.g. https://your-computer.local:8443), and tap Connect.

Step 3 — Enter the pairing PIN.
After connecting, the login screen asks for the 6‑digit pairing PIN shown in the Connect a device window on your computer. Type it in and press Pair & Continue. You only do this once per device.

Tip: If the .local address doesn’t work on your network, use the numeric address (e.g. https://192.168.x.x:8443) shown in the same window instead. Note that a numeric address can change if the computer’s IP changes.

If a device has trouble connecting (mainly older iPhones/iPads)

If scanning or connecting fails, the app reveals a “Trouble connecting? Trust Pulp’s certificate” helper. Because everything is served securely over your own network, some devices need to trust Pulp’s local certificate once:

Tap Open certificate page in Safari and install the profile it offers.
Go to Settings → General → VPN & Device Management, tap the Pulp profile, and tap Install.
Go to Settings → General → About → Certificate Trust Settings and turn ON the local certificate (it appears as “Pulp Free Chat Local CA” — on some builds it may still read “Axion Free Chat Local CA”).
If asked, tap Allow for Local Network access.

This one‑time step is also what enables the microphone (voice) on that device.

Accepting the License Agreement

The first time you install and open Pulp Free Chat, you’ll be asked to accept the End User License Agreement (EULA). Choose I Accept to continue, or Decline to stop and not install.

The current agreement is Version 1.0 (effective July 20, 2026). You can re‑read, download, or print it at any time from the About dialog.

Chatting

Type your message in the box at the bottom and press send. The assistant’s reply streams in live, word by word, as it’s generated.

While you wait, you may see:

“Thinking…” with three animated dots — the normal short pause before the reply starts.
“Loading the model — first use can take a moment” — on a genuine cold start the first time you use a model, and whenever you switch models. After that, responses start faster because the model stays warmed up.
A Stop button in place of Send while a reply is streaming. Press it to stop early; whatever was written so far is kept.

Pulp waits up to five minutes for a large model to produce its first word before giving up. Big models on laptop‑class hardware can legitimately take that long to get going, so a long pause on the first question is not a failure.

Your conversations, organized:

Each conversation is a thread, listed in the sidebar. A new thread is created automatically the first time you send a message, and it’s named after your first message.
You can put threads in folders, and rename, share, move, or delete them from the sidebar.
Your history is saved locally and survives restarts.

Good to know: Once you send a message, Pulp keeps working even if you close or background the app — the answer is finished and saved. Only pressing Stop cancels a reply.

Choosing and downloading models

Pulp can run several different AI models, and you choose which one answers.

Switching models. Click the model picker at the top of the chat. Under Your models you’ll see everything currently installed, each with a friendly name and an icon; expand a row to see details like its size and license. Pick one and it becomes active for your chats.

When you switch, Pulp unloads the model you were using and loads the new one, showing a loading state while it does. You wait once for that load, and then the new model answers. See How Pulp manages memory for why it works this way.

Adding more models. Open Manage models… to browse the full catalog. Models are grouped so you can tell what will run well on your machine:

Installed — ready to use.
Available for your machine — recommended for your hardware.
Larger · advanced — bigger, more capable models, marked with a note that they may be slow on your hardware.

Pulp checks each model against your computer’s actual memory and marks anything that cannot run on your hardware, so you don’t spend time downloading something that will never load. If you want to download one anyway — to move it elsewhere, for instance — tick “Allow large models” at the bottom of the panel. It is off by default, and the note is honest about the trade‑off: those models can be downloaded, but they will not run on this computer.

Downloading. Click Install to download a model — you’ll see live progress (for example, “installing 42%”).

One model downloads at a time. Starting several at once used to leave the whole set stuck with none finishing, so the other Install buttons are unavailable until the current download completes.
You can cancel. A Cancel button sits next to the progress figure and takes effect about a second after you press it.
A stalled download gives up rather than hanging. If a download goes quiet for a minute, Pulp stops it and tells you, so you can simply try again.

You can Remove a model later to free up space. The Manage models screen also shows a quick health check so you know the local AI service is running.

Why “signed catalog” matters: the list of downloadable models is cryptographically signed by Orange Sun. Before Pulp downloads anything, it verifies that signature, so you can trust that model files come from the vendor and haven’t been tampered with. Each model’s license and attribution information is shown alongside it.

How Pulp manages memory

AI models are large, and a model only works if it fits in your computer’s memory. Pulp handles this for you.

One model at a time, by default. When you pick a different model, Pulp unloads the previous one first and then loads the new one. You pay one loading wait on the switch, and in exchange the model you’re using always has the machine to itself. This is the default on every computer.

If you have memory to spare: tick “Keep more than one model in memory” in Manage models. It is off by default everywhere. With it on, models stay loaded together, so switching between them is instant — worth turning on if your machine has plenty of RAM, and best left alone if it doesn’t.

The memory indicator. Pulp shows whether memory is fine, tight, or full, and tells you when no model is running right now. It reflects what your computer is actually doing, including when it has started swapping to disk — the state that makes everything feel slow.

If a model genuinely can’t fit. Pulp says so plainly and suggests choosing a smaller one. It won’t tell you to close other applications in that case, because closing applications cannot change how much memory your computer has.

Keeping Pulp up to date

Open Manage updates from the header to see and control updates.

Check for updates automatically — on by default. Pulp checks when you open it and roughly every hour after that. When an update is found, the panel says Update available and the button becomes the way to get it.
Download updates automatically — off by default. With it on, an available update is fetched quietly in the background. Nothing is installed until you say so — the panel is explicit that “Downloading is your call — nothing happens until you say so.”
Install & restart — when an update is downloaded and verified, this installs it and restarts the app. Installing is always an explicit choice.
Check for updates — you can always check by hand, whether or not automatic checking is on.

Your chats and installed models are preserved across updates.

Talking to Pulp with your voice

Voice is fully on‑device: speech‑to‑text and the spoken replies are both produced locally.

You’ll find the voice controls next to the message box:

Microphone (push‑to‑talk): tap to speak, tap again (or just stop talking) to finish. Your words are transcribed into the message box so you can review or edit them before you press send. This mode does not send automatically.
Conversation mode (hands‑free): the app keeps listening and replies out loud, re‑arming the mic after each answer for a natural back‑and‑forth. In this mode your speech is sent automatically after a short pause, and spoken replies are turned on.
Speaker button: turns spoken replies on or off. Muting also stops any reply that’s currently being read aloud.
Voice settings (gear icon): choose which voice you hear. There are 14 voices (American and British, male and female); the default is “Heart.” Tap the preview next to any voice to hear a sample. You can also set how long a pause counts as “done talking” for hands‑free mode.

Attaching files

You can add files to a message in three ways: click the paperclip, paste a file, or drag and drop it onto the window (you’ll see a “Drop files to attach” overlay).

Size limit: up to 25 MB per file.
Images show as a thumbnail you can click to open; other files appear as a chip with the filename and size.

Important — what the AI can read today: the model reads text‑based files (plain text, Markdown, CSV, JSON, XML) and includes their contents in your question. Currently, images, PDFs, and Office documents (Word, Excel) are attached and shown in the chat, but the model does not read their contents — the current build ships without a document‑ or image‑reading step. Support for reading PDFs and other document types may be added in a future update. Very large text files are trimmed to keep things responsive.

Searching your chats

Open the chat search panel to find past conversations. There are two modes you can switch between without retyping:

Word Search — classic keyword search. Matching words are highlighted in a short preview of each result.
AI Search — searches by meaning, so you can find a conversation even if you don’t remember the exact words you used.

You can also sort results by Relevance or Recent.

If AI Search isn’t ready yet on your machine, Pulp automatically shows the closest word matches instead and tells you so — you’ll never get an empty screen. Your chats are encrypted at rest, and search runs privately on your own machine.

Double Check & Triple Check (second opinions)

Not sure about an answer? Every reply has a Verify menu (a shield icon, “Re‑check this response with other models”). It offers:

Double Check — Pulp keeps the answer you already have and asks one other model to weigh in, then gives you a single improved answer that takes both into account.
Triple Check — the same idea, but with two other models for a broader cross‑check.

The extra models are automatically chosen to be genuinely different from the one that gave the first answer (a different model family, not just a different size), so you get a real second opinion rather than an echo. Pulp compares the answers behind the scenes and streams back one clear, consolidated response.

Fully private: Double Check and Triple Check run entirely on your own machine with no internet access. They require at least one model from a different family to be installed — if you don’t have one yet, the option is greyed out with a note suggesting you add one.

Web search & Verify with web

Because Pulp is local by default, it doesn’t touch the internet unless you ask it to. There are two opt‑in ways to bring in fresh web information.

The “Search the web” toggle

In the message box there’s a magnifying‑glass toggle labeled Search the web. It is OFF by default.

The first time you turn it on, Pulp shows a short consent prompt (“Search the web?”) explaining that turning it on sends your question to the internet to fetch live results, that it’s off by default and applies per message, and that results come back with citations. Choose Turn on web search to enable it, or Not now to leave it off.

When web search is on, you’ll see a “Searching the web for: …” indicator, and the answer includes a Sources list — numbered, clickable links to the actual pages Pulp used.

“Verify with web”

In the same Verify menu (next to Double/Triple Check) there’s Verify with web (a globe icon). This re‑checks the current answer against fresh web sources, using a different‑family model to cross‑examine it, and shows you a corrected, cited answer with its Sources.

This is handy when an answer might be out of date — for example, a local model might not know about a recent event, and Verify with web will catch and correct it, with a link to the source.

The Sources you see are the real pages Pulp actually fetched — not links guessed by the AI — so you can click through and check them yourself.

Trash bin

Deleting a chat is safe and reversible:

Delete moves a chat to Trash (you’ll be asked to confirm — “You can restore it from Trash”). Trashed chats move into a Trash section at the bottom of your sidebar.
Restore puts a chat back where it was. You can restore one at a time or several at once.
Delete permanently (on a single chat) or Empty Trash removes chats for good — these ask for a firm confirmation because they can’t be undone.

Chats left in the Trash are automatically removed after 30 days.

Conversation memory

Pulp keeps a private, on‑device memory of your conversations so replies can stay consistent and pick up context across sessions, rather than treating every message as a blank slate. Like everything else, this memory lives on your own hardware.

On the Orange Sun AI Cube appliance, this extends to document memory — you can add your own documents (PDF, Markdown, text) and ask questions grounded in them.

The About dialog

Click the info icon in the header (About Pulp Free Chat) to see:

The app name and version (currently 1.1.24).
A Third‑Party Licenses and Model Information link that opens www.orangesun.ai/third-party-licenses in your browser, with a note that license files may also be stored alongside any models you download.
A License Agreement button that opens the full EULA, which you can Download (.txt) or Print.

Your privacy

Privacy is the whole point of Pulp Free Chat:

No accounts, no email, no sign‑up. There’s nothing to register.
No cloud, no server‑side history. Your conversations live only on your own devices and network.
The AI runs locally. Your questions are answered by models on your own hardware.
Your chats are encrypted at rest, and searching them happens privately on your machine.
Voice stays on your device — speech recognition and spoken replies are produced locally.
The only time anything leaves your network is when you explicitly opt in — by turning on Search the web or Verify with web for a message (and a visible update check). Even then, web results come back with citations so you can see exactly what was used.

What’s new since 1.1.17

If you’ve been using Pulp for a while, here’s what has changed.

Windows is now available. Pulp Free Chat ships for macOS (Apple Silicon and Intel) and Windows (x64), plus the web/LAN client for connecting other devices.

Switching models is reliable. Pulp used to try to judge whether there was enough memory at the exact moment you clicked, and it could refuse with a message like “…needs about 10.7 GB of memory. About 3.0 GB is free right now” — often for a model that would have loaded fine. It no longer works that way: it unloads the old model, loads the new one, and shows you a loading state. Related fixes along the way:

Asking a question of a model that was already loaded no longer fails because a different model was sitting in memory.
Switching models while an answer is still being written no longer interrupts anything — the answer in progress finishes normally.
When a model genuinely is too big for your computer, the message says so plainly instead of suggesting you close applications, which could never have helped.

You choose how memory is used. New “Keep more than one model in memory” setting in Manage models — off by default, for people with memory to spare.

A memory indicator tells you whether memory is fine, tight or full, and whether any model is running.

Models are checked against your hardware before you download them, with an “Allow large models” override if you want one anyway.

Automatic update checking actually works. It used to check only once a day, so the setting could look like it did nothing. Pulp now checks when you open it and roughly every hour, and optionally downloads updates in the background — while still never installing anything until you choose Install & restart.

Model downloads behave. One at a time, cancellable, and a stalled download stops and tells you instead of sitting at a percentage forever.

Longer patience on first use. Pulp waits up to five minutes for a large model’s first word, rather than giving up early.

What’s coming next

The following are in progress and not yet available — treat them as “coming soon”:

Android
iOS