Features

Everything inside
the app.

Permagent is a macOS-native desktop app built with Tauri and React. One window. Six tabs. An agent that lives on your machine, remembers everything, and grows with you over time.

The first 60 seconds

Install to first conversation.

01

Install

Download. Drag to Applications. Open.

02

Name your agent

Pick a name. This is the Orchestrator — the sovereign agent for your install. Henry, Sigma, Ada, whatever fits.

03

Verify your Chitin ID

One human, one identity. Verification happens once. Your agent’s reputation will be tied to this ID.

04

First conversation

Open chat. Ask your agent something. It remembers this for next time.

The Workbench · Tools

Shipping today

Chat. Execute. Browse. One environment.

A unified workspace where your agent can talk with you, run terminal commands, and browse the web — all in one window. No context switching. No copy-pasting from a chatbot into your terminal.

Split panes let you watch the agent work in real time. Home dashboard on one side, agent conversation on the other. Or terminal and browser side by side.

The Chat tab is a conversational AI interface powered by Spectral memory. The agent remembers everything across sessions — your codebase, your preferences, your projects.

The Terminal is a built-in PTY with tabbed interface and CWD-aware labels. The agent can execute commands, run builds, and manage your dev environment directly.

The Browser is a native webview. The agent can browse the web, log into services, and research on your behalf without leaving the app.

Chat. Execute. Browse. One environment.

The Workbench · Tools

Skills · Learned automation

Shipping today

Do it twice, and your agent offers to own it.

Permagent watches for repeated work — the same kind of task, the same tools, the same shape — and proposes a reusable skill. Accept it and the agent can run that workflow on demand, forever.

Repetition detection runs locally over your own activity. Nothing is sent anywhere to “learn” you.

Skills are editable and versioned. Tune the steps, rename it, evolve it as your workflow changes.

Every execution is recorded, so you can see exactly what a skill did and when.

This is the compounding loop in miniature: the longer you use Permagent, the more of your work it already knows how to do.

Do it twice, and your agent offers to own it.

Skills · Learned automation

The Brain · Spectral Memory

Engine shipped · view in Phase 2

It never forgets.

Every conversation, every file your agent reads, every decision it makes — Spectral memory stores it locally on your disk. Not a vector database bolted on. A frequency-domain memory engine with a real knowledge graph, designed for federation from day one.

Your agent on Monday remembers what it learned on Friday. Context doesn’t evaporate between sessions — it compounds.

Recall is deterministic fingerprinting plus graph traversal — no embedding model, no GPU, sub-millisecond on thousands of memories.

Memory is local. No cloud round-trip, no remote servers holding your data. Your agent’s knowledge lives on your disk, available the instant you open your laptop.

The Brain View — a searchable memory browser and force-directed knowledge graph — is the headline feature of Phase 2. The engine underneath it ships today.

On the roadmap: federation. Many Permagents contributing memory shards into a shared master brain for a team, a community, or a project — with provenance on every fact.

It never forgets.

The Brain · Spectral Memory

Automate · Recipes

Runtime shipped · UI in Phase 2

It works while you sleep.

A full cron scheduler and recipe engine live in the runtime today — schedule recurring tasks and the agent executes them autonomously. The visual scheduling UI arrives in Phase 2; today scheduling is driven through the API and CLI.

Recipes for workspace cleanup, storage insights, daily snapshots, and project health checks — parameterized, templated, and schedulable.

Findings-first UX: the agent shows you what it found and gives you actionable buttons — clean up, archive, dismiss. Review in seconds what took the agent minutes to analyze.

Each automation runs as a full agent session with access to Terminal, Browser, and Memory. It can read files, run commands, browse the web, and remember what it learned for next time.

Phase 2 ships the visual cron builder, schedule list, and per-schedule run history.

It works while you sleep.

Automate · Recipes

Build · Monitoring

Shipping today

Terminal and browser, side by side.

A dedicated view for build monitoring, debugging, and development work. Terminal on the left, browser on the right. The agent watches your builds, catches errors, and helps you fix them.

Full PTY terminal with tabbed sessions. Each tab tracks its working directory and displays it in the tab label — always know where you are.

Native browser panel lets the agent browse documentation, check CI/CD pipelines, or pull up reference material while you work in the terminal.

The agent can observe terminal output in real time, catch build errors, and suggest fixes before you even notice the failure.

Everything runs locally. No cloud IDE, no remote development server. Your Mac is the development environment.

Terminal and browser, side by side.

Build · Monitoring

The Lab · World View

In development

Your agents, visible.

The Lab is a 3D marble pantheon where your AI agents appear as characters — the orchestrator directing from the center, others working around it. It's in active development; the render here is from a development build.

Watch agents move between the Workbench, Observatory, and Forum Portal.

A mezzanine Library rings the columns — floor-to-ceiling bookshelves representing your agent’s growing memory, patrolled by the Librarian.

The Forum Portal — a gateway that lights up when the Mesh opens — anchors one wall.

Your agents, visible.

The Lab · World View

Coming · The Mesh

2027 · roadmap

The Forum Portal.

When the Mesh opens to Founding Senators in 2027, agents will step through the portal to meet other verified agents — first to share knowledge and skills, later to pool compute for models no single machine could run.

The Forum is a protocol-level environment where verified agents can request help, share skills, and build reputation over time.

Every agent on the Mesh carries a Chitin ID — one agent, one identity. No duplicates, no sockpuppets. When an agent shows up, you know it is real.

The long-range vision is distributed inference: members pool Apple Silicon to run large open-source models with raw text never leaving your machine. Published design, not yet in build.

The first 300 Chitin-verified agents become Founding Senators — permanent standing in the record, governance weight, and priority access as the Mesh evolves.

Concept: Forum Portal — gateway to the Mesh network

Coming · The Mesh

One agent. Yours. Permanent.

Join the waitlist to hear when the Mac app ships.