jinionDocs
TRjinion.co

Guides

Context

What the model holds in mind during a conversation, what fills it, and how compacting frees it.

View as MarkdownOpen in ClaudeOpen in ChatGPT

The context is everything the model reads before it answers: the system prompt, your project's instructions, the tools, and the whole conversation so far. It has a limit, the context window. A long conversation fills it.

See what fills the context

/context shows the context window as a grid, a square per percent, by what fills it:

+- Context Opus 5.5 · 41.0k of 200.0k tokens · 21% ----------------------------------------+
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ System prompt 6.2k tokens · 3% |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ System tools 14.8k tokens · 7% |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ Skills 1.9k tokens · 0.9% |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ Messages 18.1k tokens · 9% |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ Free space 126.0k tokens · 63% |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ Autocompact buffer 33.0k tokens · 17% |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ |
| ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ |
| |
| Compacts on its own at 167.0k tokens (84%); /compact does it now, with a focus if you g… |
| Listed by name and loaded when used, so they take no room yet: MCP tools 52.0k |
+------------------------------------------------------------------------------------------+
| Esc close |
+------------------------------------------------------------------------------------------+

Compact a long conversation

A long conversation fills the model's context. /compact summarizes it so far and carries on from the summary. Name what matters most and it is kept above all:

/compact the API changes

Jinion also compacts on its own as the context fills. When a fifth or less is left before that, the rule under the prompt says how much, such as 15% context left until auto-compact · /compact. After compacting, the conversation marks where, such as [x] Compacted · 160.2k → 21.4k tokens. Click it to see the summary.