Skip to main content

How context works

Each time SYMI works on a step, it reads your conversation, its own instructions and its tool definitions. All of this has to fit in the model’s context window. SYMI’s default model has a context window of up to about 524K tokens.

Context used

The chat box shows how much of the window the conversation uses. Hover over the bar to see the figures and a breakdown.
The headline percentage (“N% of context used”) is the full amount. The three rows are estimates, so they might not add up exactly. The indicator appears after SYMI’s first reply.

Automatic compaction

When a conversation reaches about 70% of the window, SYMI compacts it. It replaces older messages with a structured summary and keeps your most recent messages word for word. It shortens very long tool results first.
  • Compaction can happen in the middle of a long task. You do not need to do anything.
  • A “Context compacted” row is added to the chat, with a link to the summary. Select View compaction summary to read it.
  • The context indicator drops after a compaction.

Compact a conversation yourself

Type /compact in the chat box. SYMI summarizes the whole conversation so far and frees context.
  • /compact takes no instructions. Anything you type after it is ignored.
  • SYMI confirms with “Compacted N history items (~T tokens).”
  • If a reply is still running, wait for it or stop it with the stop button first.
If you see “This chat is longer than the model’s maximum context length. Use /compact or start a new chat.”, run /compact. If it keeps happening, start a new chat with /new.

Keep long conversations reliable

  • Large tool output is stored, not pasted. SYMI keeps a very large tool result as a file and works from a short preview. It can read or search the file later.
  • Work is saved as it goes. SYMI saves the conversation before each model request and before each tool runs. If a step is interrupted, SYMI treats the outcome as unknown and verifies before repeating work that could have side effects.
  • Interrupted messages. If your last message did not finish, SYMI shows “Your last message didn’t finish” and a Run it again button.
  • Titles. A new conversation is titled with the first words of your first message.

Session stats

Open Features and select Analytics to see statistics for all your SYMI conversations, whichever model they used: turns, steps, model time, tool time, time to first token, tokens per second, cache hit rate, and input and output tokens.

Session logs

Download a complete record of a conversation as a ZIP file.
  1. Open the conversation.
  2. Open Settings and, in the Session log row, select Download. Or type /export in the chat box.
  3. A dialog titled “Exporting Session” appears. When it shows “Session download started”, your browser is downloading the ZIP.
The ZIP contains the conversation’s record, the records of any sub-agents it used, and the images attached to them.
  • With no conversation open, the Download button is disabled.
  • Wait for a running reply to finish, or stop it with the stop button, before you export.
  • /export takes no path. Your browser decides where the file is saved.
  • In the 4Minds Symi CLI, /export tells you to use the browser.