v1.3.0
What's New
All new, from the inside out.
From the inference engine to the interface, we have rebuilt R.E.V.I.S. almost from scratch. Seven in-house cores work together to take it from an AI you just talk to, to an AI you can delegate work to. Every v1.x update is included at no extra cost.
At a Glance
Four big changes.
- 01 ENGINEInference without Python in betweenBuilt from scratch for the Mac, Q.E.L.V.A.R.I. runs models directly from inside the app.
- 02 AGENTS PROFrom talking to delegatingCode and documents alike, from drafting through to revision. The agent does the hands-on work.
- 03 TASKSWork continues in the backgroundM.μ.X.I.S. takes on delegated work as tasks, so your conversation never has to wait.
- 04 TERMINAL PROFrom the tools you already userevis in the terminal, MCP, and an OpenAI-compatible API. Call it from outside the app too.
Engine — Q.E.L.V.A.R.I.
Rebuilt from the engine up.
Written from scratch in Swift, C++, and Metal, the Q.E.L.V.A.R.I. inference engine runs models directly inside R.E.V.I.S., with no Python server in between.
Light and faithful
Our own quantisation recipes (Q4T, Q3T) make models smaller and lighter while keeping accuracy close to the original model's output.※1
More about quantisationDraft ahead, check in one pass
With MTP, a small draft model writes ahead and the main model checks it. Answers move forward in fewer steps.
Continue without rereading
Context the model has finished reading is encrypted and kept on your Mac, so returning conversations pick up right where they left off.
Agents PRO
From an AI you talk to, to an AI you can delegate to.
Whether it is code or documents, the agent handles the work from creation through to revision. You can choose how much to delegate in stages, from read-only to full autonomy.
Code
Runtimes for Python, Node.js, Java, Rust, Go, C#, and more are set up automatically as soon as they are needed. It writes, runs, and checks the code.
Documents
Word documents are displayed just as they look and edited with tracked changes preserved. It handles PDF editing too.
Handing over files
Attach files using the + button in the input bar or by drag and drop. Files are copied into a folder for each conversation, leaving your originals untouched.
Conversation
Change the subject, then come right back.
Each time you send a message, it tells whether you are continuing the current conversation or picking up an earlier one, switches back to the right thread, and starts thinking. You never have to switch threads yourself.
No need to repeat yourself
Return to an earlier Python question and it answers from where you left off, building on what it researched before.
Topics never get mixed up
Even as you move back and forth between work and casual chat, each conversation stays separate. Answers stay sharp however long you use it.
Readable while it is still writing
The L.A.T.R.I.O. rendering engine displays equations, tables, and code cleanly as they are being generated.
And More
More that is new.
Remembers your preferences
K.E.C.H.I. learns your preferences from your conversations. You can review what it has learnt under Personalisation in settings, search through entries, and delete them one by one.
From the terminalPRO
The revis command, MCP, and the OpenAI-compatible API can each be switched on or off individually.
Across multiple MacsPRO
F.A.L.Q. shares inference work across Macs on the same network. No central server required.
Before You Update
Before you update.
| Release | Available now |
|---|---|
| Pricing | Every v1.x update is included at no extra cost. |
| Requirements | Mac with Apple silicon, macOS 27 or later, at least 24GB of memory (48GB or more recommended).View system requirements |
| Model | Models for v1.3.0 (Q4T, Q3T) will need to be downloaded afresh. |
| Edition | File attachments, agents, sharing work across Macs, and CLI, MCP, and API access are available in the Pro edition. |
Same name.
Built differently inside.
SEVEN CORES / ONE CONSCIOUSNESS