Six agents running.
Which one is waiting on you?
Inferise Code is a native macOS control layer for Claude Code. Every project you have open sits along the bottom of one window as a colored chip, sorted so whatever needs you most is easiest to reach. Nothing pauses when you look away.
Your download is starting. Didn't start?
macOS 26.4 or later · Apple Silicon · Developer ID signed. We use your email to tell you about important updates. No spam.
The Project Status Ribbon
One glance tells you where to go next
Every project open in a window gets a chip, colored by how badly it needs you and ordered most urgent first. A question you haven't answered escalates after five minutes. A project you haven't touched in a week says so.
Status: Needs Input - 16 | Idle - 4 | Working - 3_
Five minutes to urgent
A question sitting unanswered escalates from Needs Input to Waiting! No output arrives while a session sits idle, which is exactly the case the escalation exists for.
A week to neglected
The staleness clock is seeded from a project's newest prior session, not from launch, so one you abandoned weeks ago reads as neglected the moment it reopens.
Your chip stays put
The project you're working in is pinned leftmost regardless of severity. The chip under your pointer must not move every time its session changes state.
Built for the way this actually goes
Every one of these came out of running too many agents at once and losing track.
You can't tell which agent is stuck.
The Project Status Ribbon puts every open project along the bottom of the window as a severity-colored chip, most urgent first. An unanswered question escalates to Waiting! after five minutes.
Your workflow doesn't survive closing the window.
A window is a workspace of projects, not one project. Terminal columns, pane names and sizes, the plan and the todo list all persist per project and come back on reopen.
Compaction eats the session and you lose the thread.
Every session is transcribed to SESSION.md in the project, and every agent question and decision to AGENT.md. Resume the last one with ⌘R, or let Auto-Resume do it.
You have no say in what the agent is carrying.
Edit the whole memory hierarchy Claude reads — enterprise scope down to project scope — in a built-in editor with atomic writes and rolling backups. Clear a session log when a thread has gone bad and start clean.
Your prompts are scattered across notes, gists and shell history.
A prompt library with global and per-project scopes, collapsible groups, filtering, and JSON import/export. Double-click pastes into the composer without submitting, so you read it before it sends.
Stepping away stalls everything.
Coming soon
A prompt left waiting past your threshold pushes to the phone and watch companion apps. Answer it there and the answer routes back into the exact session that asked.
Multiple project workspaces
Switch projects without stopping any of them
Click a chip and the editor swaps in place. Nothing is torn down, nothing is paused — every backgrounded project keeps its session live and keeps working while you're looking somewhere else.
Open the same project twice and it just selects the one you already have. Two sessions on one folder would mean two agents fighting over the same session file, so it doesn't happen.
- → Each window restores its own set of projects and its selection
- → ⌘N opens a fresh workspace
- → Closing a window names every project still running before it goes
Continuous session context
The history stays on your disk
As you work, the session is transcribed to .inferise/SESSION.md inside the project, segmented by prompt. Every question the agent asked and what was decided goes to AGENT.md beside it.
Come back tomorrow and Resume Session (⌘R) offers your three most recent. Turn on Auto-Resume and the last one picks itself back up as soon as the session is ready.
Prompt library
The prompt that worked, where you can find it
Keep prompts globally or scoped to one project, organised into collapsible groups you can filter. Single-click to edit, double-click to run.
Running one pastes it into the composer and stops there — deliberately. You get to read and edit before anything is sent. Import and export the whole library as JSON to move it between machines or share it with your team.
Planning & todo injection
Write the outline. It handles the rest.
Sketch a rough overview in the plan pane and send it. Inferise Code puts the agent into plan mode, frames your text into a proper planning prompt, and asks it to write the settled plan back to the file.
Then it files the result itself — agreed plans sorted to the top, live drafts beneath, superseded ones archived. The agent is asked only to append one section, which is the smallest instruction it can be given and so the one it's likeliest to follow.
Todos work in both directions: send a task into the prompt and it removes itself from the list, or capture a half-written prompt out of the input box straight into the todo list for later.
Inferise Agent
Let the safe confirmations answer themselves
When the agent stops to confirm an action, on-device intelligence classifies whether proceeding would be destructive and shows you the verdict. Nothing leaves your Mac to do it.
Opt in to acting autonomously and a confirmation judged safe is accepted for you. A destructive verdict, a plain question with no options, or no pre-selected choice all still wait for a human.
Off by default. Autonomous answering is a toggle you turn on, never something that happens to you. On-device evaluation needs a Mac with Apple Intelligence.
Answer from your phone. Or your wrist.
When a prompt has been waiting longer than your threshold, it pushes to the companion apps for iPhone and Apple Watch. Answer it there and the answer is routed back by request ID into the exact session that asked — never a different project, and nothing at all if the question already resolved.
A question you answer at your desk inside the grace window never notifies at all, and one already dealt with is withdrawn, so your phone doesn't pile up with prompts that no longer matter.
The rest of it
Terminals that keep their names
Split panes into columns and rows, name them, size them. A pane's number is assigned once and never reused — the terminal you called Terminal 3 stays Terminal 3 for as long as it lives.
Open from anywhere
Run incode . in any terminal, or right-click a folder in Finder and choose Open in Inferise Code.
Quick tools in the toolbar
Click the branch name to open your Git client, click the path to open your editor, option-click it for Finder. You choose which apps those are.
Keyboard first
⌘← navigator, ⌘→ agent drawer, ⌘↑ flip the layout, ⌘↓ snap every terminal to 80×24, ⌃⌘T new terminal.
Updates that come to you
Signed, automatic updates over Sparkle. Check for Updates is in the app menu whenever you want to look yourself.
Your files, in the open
Plans, todos and logs are plain Markdown in a .inferise/ folder in your project. Commit them, diff them, or delete them. Nothing is locked in a database.
Requirements
Inferise Code drives the Claude Code CLI — it doesn't replace it and doesn't bundle it. You bring your own install and your own subscription.
- Operating system
- macOS 26.4 or later
- Hardware
- Apple Silicon
- Also needed
- Claude Code installed and on your PATH Apple Intelligence, for on-device agent evaluation only
- Signing
- Developer ID signed, hardened runtime
- Price
- Free while in beta
- Updates
- Automatic, over Sparkle
Questions
No. It drives it. The real claude CLI runs inside the app in a real terminal, and everything you know still works. What Inferise Code adds is the layer around it: many projects at once, status at a glance, and control over the context it's carrying.
Yes. Claude Code is not bundled and Inferise Code never touches your billing — you install Claude Code yourself and sign in as you normally would. If it isn't found on first launch, the app offers to help you install it.
On your Mac. Your projects, prompts and settings are JSON files in Application Support; plans, todos and session logs are Markdown in a .inferise/ folder inside each project. On-device evaluation runs through Apple Intelligence and doesn't leave the machine. The app declares no tracking, and collects no data linked to you.
Because it's a real terminal running your real toolchain against whatever folders you point it at, and it reads the same configuration Claude Code does. That doesn't fit inside a sandbox. The build here is Developer ID signed with the hardened runtime enabled.
No. Inferise Code is Apple Silicon only and requires macOS 26.4 or later.
The app checks for updates on its own and installs signed releases through Sparkle. You can also check whenever you like from Check for Updates in the app menu.
Stop guessing which one needs you
Free while in beta. macOS 26.4 or later, Apple Silicon.
Your download is starting. Didn't start?
We use your email to tell you about important updates. No spam.