Own Your AI (OYA) — a sovereign AI assistant and agent that runs on your machine
The assistant that answers to you: runs on a model you own, counts every outside connection, and connects to every Ownware tool. Buy once, run forever.
The problem it solves
Cloud AI reads your documents, bills you per seat every month, and can change its terms or its behaviour overnight. The free local chat apps fix the privacy part and then stop at chatting: no files, no tools, no proof, no way to make it work for you.
Own Your AI is the missing piece. It is a complete assistant and agent — chat, documents, web research through a search engine you own, files, code, calendar, mail, spreadsheets, PDFs — that runs entirely against a model on your own hardware or one you choose. Every outside connection it makes is counted live in the app and written into a report you can sign and hand to a client or an auditor. And because every Ownware tool ships an MCP endpoint, OYA drives your invoicing, rostering, work orders and the rest under the same guards you face.
One purchase. Source included. A frozen core with no update race — what it is and is not is written down, and it stays that way.
What you get
Runs on a model you own
Any OpenAI-compatible local server — llama.cpp, Ollama, LM Studio, vLLM. Recommends a model that fits your hardware and downloads it, verified. Or point it at an endpoint you choose; the app then says REMOTE, in red, because that is what it is.
The ledger: every outside connection, counted
The OUT channel shows how many connections left your machine this session and to which hosts — live. Airplane mode blocks everything; sealed mode runs shell and scripts with no network so the count is the whole number. Prove it → a signed report you can hand to anyone.
A real agent, with the trace on the table
Plans, uses tools and acts — files, shell, code, search, pages, PDFs, spreadsheets, calendar, mail, SQL — with a live one-line status while it works and the full trace afterwards. Mutating actions ask first, unless you say otherwise.
58 tools in the core, and pieces you choose
Read, write, diff, search files; run tests; import CSV/XLSX into SQL; read and write .docx/.xlsx/PDF; redact personal data; calendar .ics; IMAP/SMTP mail over TLS; a private web search through your own SearXNG. Then add pieces — MCP servers, plugins, skills — each showing where its data goes.
Connects to every Ownware tool
Every product in this store ships /mcp. Paste an API key from the product's API-settings page and OYA reads your invoices, rosters and work orders and answers in plain words — walked end to end against live products, read-only, with the transcripts.
Web app and terminal, one engine
A first-class browser app (the Mixer: sessions, pieces, files, knowledge, skills, ledger, prove-it) and a full CLI with sessions, scheduling, blueprints, cron and a signed attestation chain. Both talk to the same local engine.
Frozen core, no update race
The core is finished software: SCOPE-FREEZE.md fixes what it is; 830+ tests keep it there; security fixes and honest maintenance only. You will not wake up to a different product.
Source included, yours to keep
Vanilla Node.js and ES modules, no build step, no third-party runtime dependencies (the supply chain is derived and stated in the report). Read it, change it, run it for as long as you like.
Pricing — one-time, yours forever
- ✔ Full source included — PHP you can read and change
- ✔ No subscription. One payment, yours forever
- ✔ It never phones home
- ✔ Your data stays on your server
No reviews yet — so here is what we offer instead. Every competitor price on our comparison pages is source-linked and dated, and when a vendor’s price won’t load we say so rather than guess. See the sources for yourself →
Screenshots
Honest limitations
- Not a model: you bring the model server (llama.cpp, Ollama, LM Studio, vLLM); the setup recommends and downloads one, verified
- Linux and macOS are the execution hosts; Windows users run it under WSL (the isolation verdict is self-verifying there)
- No desktop app, editor plugin, image generation, speech, browser automation or messenger bridges in the core — those are pieces, or not at all
- The public showcase at oya.ownware.io replays recorded runs; there is no model on that box — OYA runs on yours
- Mail is IMAP/SMTP over TLS with app passwords; OAuth-only mailboxes are not supported
Node.js 22+; one command (`oya up`) starts the engine and the web app; a local model server of your choice; Linux/macOS (Windows via WSL).
Frequently asked questions
What do I need to run it?
A machine with Node.js 22+ and a local model server (llama.cpp, Ollama, LM Studio or vLLM). A 24 GB GPU runs a strong 27B model; 8–16 GB runs the smaller ones the setup recommends. Linux and macOS are the execution hosts; Windows runs it under WSL.
Does it phone home?
No. The only connection made on your behalf without asking is to the model endpoint you configure. Everything else is a tool you invoke, counted in the ledger. The report lists every host that appears in the source and its stated purpose; the tests fail if a host is unclassified.
Is a model included?
No — models are licensed by their publishers and you pick the one that fits your hardware. The setup recommends one and downloads it, verified. You can also point OYA at any OpenAI-compatible endpoint you trust; the app then labels the model REMOTE.
How is this different from a chat UI like Open WebUI or a cloud assistant?
It is an agent with proof, not a chat window: it plans, uses tools and acts on your files, mail, calendar and connected software — and every outside connection is counted and reportable. Cloud assistants run on someone else's machine on a monthly seat; local chat UIs stop at chatting.
Does it work with my Ownware apps?
Yes — every Ownware tool ships /mcp with a bearer key from its API-settings page. Add it as a piece (or in ~/.oya/config.json) and OYA drives it under the same roles and approval guards the product enforces. The recipe is in the docs, and it was verified against live products.
Will it get updates?
The core is frozen on purpose: security fixes and honest maintenance, no feature churn. What you buy is what it stays. Pieces — MCP servers, plugins, skills, data packs — are where growth happens, and each says where its data goes.
Can I use it for clients or in my company?
One purchase covers your own machines (personal) or your organisation's internal use, source included; no redistribution or reselling. Reports and attestations are evidence, not certification.