SYSTEM TRAY · WINDOWS · AGENT-CALLABLE
TrayTool is a capability marketplace that lives in your system tray. Independent developers ship features as standard plugin packages — keeping full ownership of IP, pricing, and distribution. You collect what you need, or install a ready-made Edition. And every capability you collect is built to become a tool your AI can invoke.
Good evening. What are we making?
01 THE GAP
VibeCoding drove the cost of building a working tool toward zero — but distribution and invocation never moved. Five platforms, five review queues, five privacy policies. And the tools still can't talk to each other, or be discovered by any AI. The bottleneck has moved from production to distribution — and to being called.
THE BREAK CAME IN 2024–25. THE CURVES STILL HAVEN'T RE-CROSSED.
USERS
You end up as your own integrator, with a clipboard for a bus. The capabilities are on your machine — the workflow isn't.
“My computer has capabilities. It doesn't have a workflow.”
DEVELOPERS
You end up as your own publisher, with a timeline for a channel. The code is good — and it sits in the dark.
“I built something good. Getting it seen is a different job — one nobody pays for.”
02 ARCHITECTURE
Supply, assemble, invoke. The first two exist elsewhere, in pieces. The third is the moat: no one else makes capabilities agent-callable the moment you collect them.
SUPPLY01
Package features as signed .tray plugins — one manifest, one command plane. IP, pricing, channels, and update cadence stay with the person who wrote the code. Three distribution channels, including one where we take 0% and stay out of the way.
ASSEMBLE02
Collect plugins like cards, or install a ready-made Edition that reshapes the host around a single identity. Your first screen becomes exactly what you need — and nothing you don't.
INVOKE03
Every collected capability is designed to surface on one unified command plane — exposed to agents through MCP, gated by a three-tier security gate, written to the audit log. A plugin that only has a UI is an incomplete plugin here.
THE MOAT — SHIPS IN PHASE P203 INTERFACE
A first-time user and a power user shouldn't see the same screen — but they can share the same product. Not two products: two views of one workflow.
What do you want to do?
Split this outline into a 12-episode storyboard
Same orchestration template. The interface decides how much you see.
04 EDITIONS
An Edition is a declarative manifest — which plugins, which defaults, which first screen, which words. Install one, and the host becomes software made for someone like you. Sharing an Edition shares the whole list: one link offers to install every missing plugin.
Outline to storyboard to voice to asset pack — a desk built for one-person studios.
“Split this outline into a 12-episode storyboard.”
2 plugins live · 2 planned
Bedtime stories, gentle timers, a calmer evening routine.
“Tell a bedtime story about the North Pole.”
2 plugins live · 2 planned
Fade-out playback on a dark, quiet screen. Nothing else.
“Stop in forty minutes.”
1 plugin live · 2 planned
Editions are the start, not the cage — add any plugin, restore any hidden panel, exit anytime. Editions ship in Phase P3; share links like traytool.com/w/you carry them.
05 SUPER-CHAINS
A super-chain is an ordered composition across plugins. The value of n capabilities that can call each other grows toward O(n²) — which is why connection beats feature count.
We sequence chains by smallest gap first — mechanisms before harvest. Parenting and Sleep validate the loop; Film and E-commerce monetize it.
07 STATUS — MEASURED, NOT MARKETED
We don't dress the roadmap up as the present. Every number below is measured in the repository on a stated date — and every target is labeled with the phase that delivers it.
| DIMENSION | MEASURED TODAY | DESIGNED TARGET |
|---|---|---|
| Host codebase | 234,980 lines C# · 22 modules · 512 files | — |
| CLI command plane | 71 commands · 15 domains | — |
| Plugin business commands | 14 — OWMap only; the other 5 plugins ship 0 | 60+ across all plugins |
| Security & audit | 3-tier risk gate + audit pipeline | same |
| Built-in plugins | 6 | — |
| Third-party plugins | 0 | ≥ 30 |
| Agent-visible tools | 0 via MCP — server not built | ≥ 40 via MCP |
| Editions | 0 | ≥ 12 |
| Marketplace & payments | not built | Phase P5 |
THE ROUTE
P0 · Contract freeze
.tray format, permission model, compatibility promise
P1 · Runtime loading
Plugins load without recompiling; SDK & scaffold
P2 · Capability plane
MCP server live — agents can invoke the desktop
P3 · Editions & channels
Edition manifests + three distribution channels
P4 · Launcher
Standalone edition shells for vertical users
P5 · Marketplace
Discovery, escrowed bounties, payments
“A product that honestly says ‘we just started — the foundation is solid, the route is clear’ is more persuasive to developers than one with suspicious numbers.” TT2-01 · NARRATIVE DISCIPLINE
08 PRICING
The line is drawn by one test: whether the platform pays real, ongoing costs. Everything local stays free — structurally, not as a trial.
ALWAYS FREE
Everyone — collectors, configurators, developers
$0
FREE IS STRUCTURAL — NOT A TRIAL
PRO
Heavy users of the sprite & workflows
$—
PRICED AT LAUNCH · PHASE P2+
TEAM / ENTERPRISE
Organizations with boundaries to keep
Talk to us
SALES FIRST · SELF-SERVE LATER
Plugin prices belong to their developers — the host will never charge rent on installing them.
09 QUESTIONS
Structurally, yes. The host, the plugin contract, and all three distribution channels stay free — no feature gates, no plugin-count limits, no invocation metering. We charge only where we pay real, ongoing costs: cloud compute, cross-device sync, enterprise isolation. Free is the premise of the marketplace, not a trial.
Windows only, today. The host is a native WPF tray application — resident, light, and honest about what it is. Cross-platform is a form-factor decision we haven't made, and we won't announce it until it's real.
Honestly. .NET has no in-process code sandbox, so we don't pretend otherwise. Security is tiered: unsigned plugins are structurally denied high-risk permissions; every invocation passes a three-tier risk gate and lands in the audit log; out-of-process plugins get real OS-level isolation — and the UI tells you which is which, in plain words.
The command plane and the security gate already run in production. The MCP server that exposes them to agents ships in Phase P2. Until then, any claim that “your AI can call everything on this machine” would be exactly the kind of promise we refuse to make.
None, structurally. Self-distribution is 0% forever, and we don't punish anyone for using it — if anything, we track its adoption as proof that our definition-rights promise is real. The official payment channel takes 5–8% (planned), sized to cover payment, dispute, and compliance costs. Not a profit center, by design.
No. A launcher points at shortcuts; TrayTool is a capability marketplace with a plugin contract behind it — signing and trust tiers, a three-tier risk gate with an audit pipeline, and one command plane (71 commands across 15 domains today) that agents will invoke over MCP once that server ships. What the tray-launcher category never modeled — third-party developers owning their IP, pricing, and distribution — is the entire premise here.
No cookies, ever. The site stores only two localStorage entries (tt.lang and tt.geo) to remember your language and, on a first visit whose browser language is ambiguous, a two-letter region code fetched from geojs.io (fallback ipwho.is) purely to pick a readable language — it is never sent anywhere, and your manual choice always wins. There are no analytics scripts, ads, tracking pixels, fingerprinting, or forms. The one server-side element is the download counter (dl.php): when you download, it logs a timestamp, IP address, user-agent, and referrer to count downloads and spot abuse. That is a server log, not a cookie. Chapter 3 of the legal page lists every item and marks which lines you can verify in the source.
Measured on 2026-09-07, in the repository: a 234,980-line C# WPF host, a 71-command CLI across 15 domains, 6 built-in plugins, and a live three-tier risk gate with audit pipeline. Not built yet: the third-party plugin ecosystem (0 today), the MCP server agents will call (Phase P2), Editions, and the marketplace with payments (Phase P5). The measured column and the designed-target column sit side by side in the status section above — you never have to guess which claim is which.
No — and today it can't do anything at all: the MCP server that would let agents call the desktop is not built yet (Phase P2). When it ships, agent invocations go through the same gate everything else does: the three-tier risk check, the audit log, and the trust tier of the plugin involved — unsigned plugins are structurally denied high-risk permissions. The design goal is that “my AI called my desktop” is always an event you can inspect, not a leap of faith.
One host in your tray. The capabilities you chose. An AI that can finally act on them — soon, and honestly labeled until then.