FAQ
Frequently asked questions.
Straight answers about what Build Perch is, what leaves your machine (and what never does), and how the trial and licenses work.
The basics
What is Build Perch?
A local-first AI workspace for Windows 11. It puts every major AI model (GPT, Claude, Gemini, Grok, and hundreds more via OpenRouter, or fully offline models via Ollama) into one desktop app that works directly on your files: documents, spreadsheets, presentations, task boards, media, and scheduled automations. Your files and chats stay on your machine.
What do I need to get started?
Windows 11 and one model source: an OpenRouter API key (pay-as-you-go access to the cloud catalog), a free local Ollama install, or an Ollama Cloud subscription. Connect any one and you're chatting. The 14-day trial needs no credit card.
Does it work offline?
Local models through Ollama work fully offline, and so do the built-in local features: semantic search, voice dictation, and media editing. Cloud models need a connection, and starting a trial or activating a license needs one; after activation the app runs up to 30 days offline.
Which platforms are supported?
Windows 11, 64-bit (x64). Windows on Arm isn't supported.
What are the system requirements?
A 4-core x64 CPU (any chip that runs Windows 11 qualifies), 8 GB RAM (16 GB if you run local models), and about 1 GB of free disk. The local speech, embedding, and media models ship inside the installer, so there are no downloads on first run. No graphics card is required by Build Perch itself. For heavier on-device work (local transcription, indexing big workspaces, local video editing), a 6-core Intel Core i5 (10th-gen or newer) or comparable Ryzen is smoother. Full details are on the download page.
Models & cost
Which AI models can I use?
The live OpenRouter catalog: 400+ models across every major lab, with new ones appearing the day OpenRouter lists them, no app update needed. You can also run any model you've pulled into local Ollama, plus Ollama Cloud's hosted open models. The picker shows what each model can do (vision, tools, media) and what it costs per token.
Do I pay Build Perch for AI usage?
No. You bring your own provider accounts and pay them for your own usage. Build Perch takes no cut of your AI spend and never sits between you and your provider's billing. Local Ollama models cost nothing to run.
Can I switch models mid-project?
Yes. Each chat is pinned to a model but switchable any time, and every reply records exactly which model produced it, so a cheap model can do the legwork and a frontier model the final pass, in the same workspace.
Can I run everything locally?
Chat, search, dictation, transcription, and media editing, yes. With Ollama models, nothing leaves your machine at all. The one exception is media generation (images, speech, music, video), which runs through OpenRouter's hosted media models.
What is Ollama Cloud?
Your ollama.com subscription's hosted models, connected as a third provider: a flat-rate way to run bigger open models without a local GPU.
Privacy & security
Where is my data stored?
On your machine, as ordinary files you can see. A workspace is just a folder; chats, task boards, generated documents, and the search index all live inside it. There is no Build Perch cloud holding your content.
Can Build Perch (the company) see my files or chats?
No. There's no telemetry or analytics, and no Build Perch server in the data path: your content goes only to the AI provider you chose, under your own credentials. The app itself calls out only for license and trial checks (a license key and a one-way-hashed machine fingerprint, never content), anonymous update checks, and (only if you ask for it) a one-time re-download of its bundled speech model. None of them ever carry your content.
What actually leaves my machine?
Two things: the requests you send to your chosen AI provider, and outbound actions you authorized (an email, a Discord post, a Notion write). In chat, every send stops at an approval banner showing exactly what's going where; sends never auto-approve, even with auto-approval turned all the way up. In unattended automations, sends go only to destinations you pre-authorized when you set the agent up. A recipient can never come from message content, and anything not on your list is held for your approval instead of sent.
What is Privacy Mode?
On by default: every request the app makes to OpenRouter (including media generation and background summarization) carries a zero-data-retention flag, so OpenRouter routes only to providers that attest they don't store or train on your prompts and files. It's their attested policy enforced by OpenRouter's routing, not a guarantee we make on their behalf, and the AI's own tools can't reach the toggle.
Can I block the AI from specific folders?
Yes. Right-click any folder and lock it. The lock is enforced in the app's native core on every read, write, move, delete, listing, and even semantic-search result. The AI can't read, write, or unlock it, no matter what it's told.
What if a malicious email or webpage tries to trick the AI?
Content that arrives from outside (emails, Discord messages, watched files, fetched pages) is treated as untrusted: it's fenced with unforgeable markers so the AI treats it as quoted material rather than instructions; URL-fetch tools are withheld for that run; and replies can only go back to the sender. Automation runs can additionally be confined to a single input folder and a single output folder.
Are my API keys safe?
They live in the Windows Credential Manager, not in app files. Integration secrets (Gmail, Discord, Notion, GitHub, and others) are write-once: the app's UI can store or replace them but can never read them back; only the native backend uses them, against a fixed allowlist.
Are updates safe?
Update packages are cryptographically signed and verified on your machine before they install, from a fixed first-party feed. Updates never read or carry your content.
Trial, pricing & licensing
How does the free trial work?
One click on first launch starts a 14-day trial with the full feature set: no credit card and no feature-limited tier. The trial is tied to your machine, so reinstalling doesn't reset it.
What happens when the trial ends?
The app asks for a license at the next launch. Nothing is deleted: your workspaces, files, and chats stay on disk exactly where you left them, and they're all there the moment you enter a key.
What's the difference between the subscription and the one-time purchase?
Identical features, no differences at all. The subscription runs while you're subscribed. The one-time license never stops working; it includes 12 months of updates, and after that you can keep using your version forever or renew to keep receiving new ones. Current prices are on the pricing page.
How many computers can I use it on?
Paid licenses activate on up to 3 machines; the trial on 1.
What if I'm offline for a while?
A signed local pass keeps the app fully working for up to 30 days without a connection. Perpetual licenses never lock the app regardless: expiry only pauses updates.
What information does licensing send?
Your license key and a machine fingerprint (a one-way hash of your machine ID; the raw ID never leaves your device), plus the OS name and a generic device label (never your computer's name). Starting a trial sends only the fingerprint and app version. Your files, chats, and prompts are never transmitted.
Do you store my payment details?
No. Purchases are handled on the website by Stripe, our payment provider and merchant of record; the app itself never sees payment information; it only takes a license key.
What's the refund policy?
See the refund policy page; it's short and plain-English.
Using the AI
What can the AI actually do?
It calls 100 built-in tools: read and search your files, research the web, create and edit Office documents, manage your task board, generate and edit media, remember past work, and run saved automations. Every tool call shows up as an expandable activity row: what it's doing, with what arguments, and how it went.
Do I have to approve every single action?
You choose. Auto-approval has tiers (always ask, approve reads only, or approve everything) and it resets to safe defaults each time the app starts. In chat, anything that leaves your machine always asks, regardless of the tier; automations instead send only where you pre-authorized, and hold anything else for your review.
Which file formats does it handle?
It creates Word, Excel, PowerPoint, PDF, HTML, and SVG, and edits them all in place except PDF. Office edits are surgical, preserving themes, charts, and masters. It reads 75+ file types, including code, text, and Office formats, and its local search indexes dozens of them.
Can it edit video and audio? Does my footage get uploaded?
Yes, and no. Ten editing tools (trim, join, split, reframe to vertical, clean audio, extract audio, burn captions, transcribe, and more) run entirely on your machine through a bundled FFmpeg. Your footage never uploads anywhere.
Can it generate images and video?
Yes: images, speech, music, and video, right in chat, through OpenRouter media models (including image-to-image and image-to-video). Generation is the one media feature that requires a cloud model; editing stays local.
Is there voice input?
Yes. The mic button transcribes speech on-device with local Whisper. No audio ever leaves your machine.
Does it remember things between chats?
Two ways, both inspectable. Workspace memory is an opt-in, visible markdown file the AI curates about that workspace; you can open, edit, or revert it. And the AI can search your other chats in the workspace when asked. Long conversations are summarized in the background with a live meter, so context is never silently dropped.
Can I automate recurring work?
Yes. Save agents (a standing prompt, model, and tool allow-list, with an optional validator model that can veto bad tool calls) and chain them into workflows that run on triggers: a schedule, an incoming email, a Discord message, a file change, or another workflow finishing. Every run gets a History card with one-click Revert that restores every file the run touched.
Which integrations are supported? Can I add my own plug-ins?
Gmail (as a dedicated mailbox you forward to, over standard IMAP/SMTP with an app password), Discord and Notion (their official APIs), and GitHub (over MCP, running GitHub's own first-party server), plus web search via DuckDuckGo and Brave, Wikipedia, arXiv paper search, and Openverse image search. There's deliberately no arbitrary plug-in system: every integration is curated and in-house, so no third-party plug-in code ever touches your local data.
Practical bits
Where does the AI's work get saved?
In your workspace, as ordinary files: generated documents, images, audio, and video land in a visible BuildPerch folder (or wherever you tell it to save). Open them in any app; nothing is trapped in a proprietary store.
Can I undo what the AI did?
Yes, at three levels: each AI reply's file edits can be reverted from the message itself; each automation run has a one-click Revert that restores every file it touched; and AI file deletions go to the Recycle Bin, so you can restore them.
Will it slow my computer down?
No. The app itself is light. Built-in AI features (dictation, transcription, search indexing) run short CPU bursts and never need a GPU. Only running large local models through Ollama demands serious hardware, and that's entirely optional.
How big is the install?
About 420 MB, everything included: the local speech-to-text and search models ship in the installer, so there's nothing to download on first run.
Didn't find your question? Email [email protected].