Dashboard
The dashboard is the first screen in onplate. It answers three questions: what is installed into each AI tool, what needs attention right now and what you actually use.
The AI system map
The main block. A table with a row per AI tool — Claude, Codex, Cursor, GLM and OpenCode — and these columns:
| Column | What it shows |
|---|---|
| AI tool | the name and how many resources were found |
| Context | the approximate standing weight in tokens — “always in context” |
| Skills / MCP / Agents | how many resources of each kind onplate manages |
| Status | “all good”, or the list of problems: threats, conflicts, external copies, outdated installs |
Above the table are filters by resource kind and a summary: when the map was built, how many resources there are in total and how many need attention. The state of the data is labelled honestly: up to date, may be stale or the check did not finish.
Clicking a row opens a window with the managed resources and the problems found.
From it you can jump straight to filtered Bindings or to
reviewing the problems. It closes with the cross, Esc or a click on the dimmed
area.
The Check system button (Check again once a check has run) reads your local AI folders and configs and rebuilds the map. It installs and deletes nothing — which is stated right under the table.
About context weight
Every skill has short metadata — its name and description. That goes into every AI session, always, before your first question. The full body of the skill is loaded only when the skill fires. The “Context” column shows exactly the standing part; the per-tool details also show the ceiling, “up to N if everything activates” — a maximum, not a forecast for a normal session.
The estimate is approximate (roughly 4 characters per token), but it is enough to notice that storage has grown fat and it is time to clear it out.
Needs attention
A list of actual chores rather than counters: conflicts among discovered skills, dangerous skills and subagents, suspicious entries, stale exceptions, packages that need a look. Every row leads to the right section with the filter already applied.
The first few entries are shown, the rest behind Show more. When there is nothing to do the block does not appear at all: an empty “no urgent actions” should not take up room on the first screen.
Usage
Statistics for the last 30 days: which skills fired and which were never used in the period. It helps you find ballast — things that take up context but give nothing back.
Firings are counted from Claude Code, Codex, Cursor, OpenCode and GLM
sessions: activation through the Skill tool and slash commands, reading SKILL.md,
$ mentions in Codex, the skill tool in OpenCode. For Cursor only a read of the
skill file itself counts: the list of available skills reaches the model with
every request, and counting that would record a mere “saw it” as work. The block heading lists whose logs
were readable. The analysis is local only: the logs give the fact — which skill,
when, in which project — never the contents of your conversations.
Resources
The summary table at the bottom: per resource kind, how many entries are in storage and how many are in AI tools. Both numbers are clickable and lead to the matching section. If a kind is installed nowhere, a Configure link appears next to it.
What can appear at the top
Cards can show up above the map — but only when there is a reason:
- an onplate update — a new version is out;
- a project profile is waiting — a
.onplate/profile.jsonappeared or changed in one of the folders onplate knows (see Project profile); - allow watching the local AI environment — a request for read access to your local folders. Reading only; nothing is sent anywhere.
If storage is empty
Instead of statistics the dashboard shows a welcome with three starting actions: Import a package, Connect a repository, Scan the system. Which one to pick is covered in the Quick start.