🚀 Naseem launched on Product Hunt — join the community forum: questions, wins, what's next →

Naseem · نسيم

The Mac-native AI agent that does the work, not just the chat.

A fast, lightweight agent for macOS — a gentle breeze against the heavy, hot Electron assistants. Chat with any major model or fully on-device, and let it safely act on your Mac.

⬇ Download for macOS Get Naseem Pro

Free plan · No account · No subscription · Bring your own AI key

Naseem is a Mac app (macOS 14+). On Windows or a phone? Follow Naseem on Product Hunt for updates, or send yourself the link for later.

Naseem Pro: $49$29 lifetime — launch price ends September 30

Naseem - A native AI agent that does real work on your Mac | Product Hunt

Naseem in action — building, launching, and driving a real app in the iOS Simulator.

Watch Naseem fan one request out to three sub-agents in parallel, then compose the results — while its own context stays tiny.

Naseem building, installing, launching and driving a real app in the iOS Simulator — sim_build_install, sim_launch, sim_ui and screenshots in the transcript

Debug your app by asking: Naseem builds, installs, launches, taps through the UI, and screenshots the iOS Simulator — natively.

An agent, not just a chat window

A powerful model alone isn't an agent — it needs a harness. Naseem is that harness, plus one thing browser-bound assistants don't have: your Mac.

LLM + Agent Harness + Your Mac

The model provides the intelligence. The harness gives it agency. Your Mac gives it the power to act.

Why Naseem is different

Most AI assistants stop at generating an answer. Naseem is built to continue — inspect the project, make the change, run the tools, read the results, verify, and fix.

It works with your real files

Inspect, read, patch, and create files on disk. No copy-pasting between a chat window and your editor.

It has a real terminal

git, xcodebuild, package managers, tests, scripts, Python. Not a browser sandbox — your Mac's actual environment.

It drives the iOS Simulator natively

The way the best browser agents work: it reads the app's accessibility tree as text and acts on named elements — not by screenshotting and guessing. Faster, cheaper, and far more precise. Other coding agents shell out to simctl one command at a time; Naseem keeps one live connection and talks to your app.

It automates native Mac apps

The same tree-first approach through Apple's Accessibility APIs — for GUI-only apps (Notes, Mail, System Settings) that have no CLI or API.

It proves, it doesn't assert

A "done" is a claim to check: a reply that says it created a file is sent back if the file isn't on disk, a Simulator tap on a screen that has changed is refused, and a scheduled rule that ran no check is refused too. The menu bar shows when the last rule ran; a status file on disk says the same. The app is the status page — no server to trust.

Its prompt stays bounded, however long the task

Finished tool work is folded out of the live context right away, so a conversation with hundreds of steps still runs on a small prompt — and your own Usage & Cost window shows it. How it works.

Nothing is claimed before it ships

Feedback from the forum has become releases within hours, several times. Every release is documented, and the philosophy page is written so that a principle the app doesn't keep counts as a bug.

It's the conductor, not the orchestra

Written entirely in Swift — no Electron, no embedded browser, no Node runtime. During a full build-and-test run it stays light while Xcode, the Simulator, and the model do the heavy lifting.

Naseem next to the iOS Simulator — the agent's debugging summary on the left, the finished bilingual weather app it built showing live Cairo weather on the right

A real session: Naseem built and debugged this bilingual weather app, then verified it live in the Simulator.

What's New

Naseem updates itself, and every release is documented. See what's new.

Naseem's Usage window — per-model token and cost breakdown, and the Memory panel showing a 174 MB live footprint with attribution

Tokens and cost per model, and the app's real footprint — measured, not promised.

Built to stay sharp on long tasks

The hard part of an agent isn't answering — it's staying useful over hundreds of steps without drowning in its own context or losing the plot.

Small, stable, cache-friendly context

Completed tool outputs collapse to compact receipts (full text always retrievable on demand), and the request prefix stays byte-identical between steps — so your provider's prompt cache does the work. Cheaper, faster long sessions; on a subscription your quota lasts far longer.

Memory that survives restarts

Build logs, test runs, and search results persist across app launches. Ask about yesterday's failed build and the agent pulls up the actual log.

Stays on mission

Each conversation carries an objective the agent re-reads on every step, with progress checkpoints so it stops grinding a stubborn subproblem and reports back.

Crash-safe runs

Quit or crash mid-task and completed steps are preserved and clearly marked; continue resumes with memory of what was done.

Naseem's Lessons panel — durable lessons the agent captured from failed or corrected runs, awaiting your review

Naseem captures lessons from failed or corrected runs — you review what it keeps.

Everything in the box

Free is the agent you chat with and supervise. Pro is the agent you hand work to.

Profiles
Per-conversation personas — Assistant, Writer, Software Engineer, QA, Automation — each with its own tools and approval policy. All five on Free; only the Simulator and Mac-driving capabilities inside them are Pro.
iOS Simulator driving PRO
Boot, build, install, tap/type/swipe through the real UI tree, read logs, screenshot. Debug your app by asking.
Mac computer use PRO
Drive GUI-only Mac apps natively via Accessibility — allowlisted per conversation, approved per action.
Sub-agent delegation PRO
Spawn focused child agents for big side-tasks and work in parallel, with live nested traces.
Telegram remote PRO
Talk to your agent from your phone — including remote Approve/Deny of gated tool calls.
Web search
A built-in web_search tool for current events, docs, and prices — bring your own free Tavily or Brave key.
On-device inference (MLX)
Fully local Apple-Silicon inference — no network, no cloud, private by construction.
Native tools
Real terminal, filesystem with diff-based patching, Python, PDF — behind per-call approval.
MCP
Connect any Model Context Protocol server; paste-whole-JSON config editor.
Memory & skills
Cross-conversation memory; recurring tasks become reusable skills; lessons learned from failed runs.
Rules PRO BETA
Scheduled agent runs that stay quiet until something needs you, refuse to say "all clear" without running the check, and send a weekly heartbeat with real run counts. The menu bar shows the age of the last run; a status file mirrors it.
Start cards & project notes
A new chat opens with six tasks Naseem can do right now — in a project, drawn from notes it keeps in naseem/ (project, roadmap, journal) that you can read, edit, and commit.
Autonomy per chat
Ask-first or run-unattended, decided per conversation from a badge in the header. Profiles can only lock it down, never loosen it.
Vision for any model
A text-only chat model still inspects screenshots: your chosen vision model describes them, and the agent treats that as its own eyes.
OpenRouter & every provider
Anthropic, OpenAI, Gemini, OpenRouter (hundreds of models, one key), Ollama Cloud, local Ollama, on-device MLX. Prices and capabilities from the providers' own catalogs.
Usage & Cost BETA
Measured tokens per call, live prices, and a memory footprint panel — with a helper model doing background chores so your chat model isn't wasted on them.
Naseem's Skills library — reusable procedures like project-builder and github-workflow, importable and reviewable

Skills teach Naseem how you want recurring work done — reviewed before you trust them.

Free to start. Pro if it earns it.

Because Naseem uses your own models and API keys, buying Pro doesn't lock you into another monthly AI bill.

Free

$0 no card
  • Every provider (Anthropic, OpenAI, Gemini, OpenRouter, Ollama, local MLX)
  • Conversations & up to 3 projects
  • Web search, vision, artifacts
  • Core tools with approval, MCP
  • Memory & skills, Quick Ask
Download Free

No account. No card.

Pro

$49$29 lifetime
Launch price — ends September 30
  • iOS Simulator automation
  • Native Mac computer use
  • Sub-agent delegation
  • Telegram remote access
  • Self-improving skills & custom profiles
  • Unlimited projects
  • Personal license, up to 3 Macs
Get Naseem Pro

14-day Pro trial · No credit card · No subscription. Keeps working on Free afterward.

The Free feature set may change in future versions. A version you already have never loses features it shipped with.

Naseem's Profiles settings — built-in Assistant, Writer, Software Engineer, QA Engineer and Automation/Ops profiles, with custom profiles available in Pro

All five built-in profiles in Free; the Simulator and Mac-driving capabilities inside them are Pro. Duplicate and customize your own with Pro.

Get started

Requirements: macOS 14 or later, Apple Silicon or Intel. Bring an API key (OpenAI / Anthropic / Gemini), or run Ollama, or nothing at all with on-device MLX. iOS Simulator automation requires Xcode. The app is Developer ID-signed, notarized, and updates itself.

⬇ Download for macOS What's new