[Bug]: interface becomes slow and unresponsive after several iterations

[bartman]

Bug Description

This may be related to #52 or #66, but not quite the same.

$ grok --version
1.0.1

installed in a full Debian/testing container...

$ sudo apt install npm nodejs
$ npm install -g bun
$ bun add -g @vibe-kit/grok-cli@0.0.27

I'm not using the latest version because #79 / #80 issue.
UPDATE: after my workaround (#80 (comment)) I'm able to run latest version -- same issue.

Everything seems to work at first. I can ask grok to build and modify code, run tests. After about 15 minutes of use, every time, the interface will grind to a halt, both typing and responses back from grok are sluggish.

If I attempt to plow through eventually I'll get to a state where I can type and watch my characters show up 3 seconds later.

If I restart the grok app, everything is back to working -- for a few prompts.

UPDATE: /clear makes it more responsive again. So it's just some state buildup.

Steps to reproduce

1. open an existing code base.
2. tell grok to scan the code base (to build context)
3. tell grok to make changes (like refactor something)
4. try to type -- notice it's slow.

[dancegit]

I also have slow behavior on the newest grok, I basically asked it to analyze the codebase and output the results to a file, and it stopped before outputting to the file, then i asked it to write the file, and it started with some <xai: function_call_name="create_file"> ... but extreeeeemly slow .. like one line every 10 seconds or so.

[zfifteen]

The behavior in this issue is caused by unbounded session state inside the CLI.

• Internally, GrokAgent keeps a growing chatHistory: ChatEntry[] plus a parallel messages: GrokMessage[] conversation log, and the UI uses that full history to drive rendering and token counting on every update. (deepwiki.com (https://deepwiki.com/superagent-ai/grok-cli/3.3-ai-integration?utm_source=openai))

• The message-display layer (ChatHistory, diff renderer, token counter / loading spinner) repeatedly walks and renders this entire history and re-estimates tokens on a 250 ms timer, which becomes increasingly expensive as the session accumulates many tool results, file contents, and diffs. (deepwiki.com (https://deepwiki.com/superagent-ai/grok-cli/6.3-message-display))

• Because there is no automatic pruning/compaction of old entries, the cost of each keystroke and each timer tick grows over time, eventually saturating the Node/Ink event loop so that even typing becomes laggy. This matches the report that the UI is fine at first, then degrades badly after ~15 minutes. (github.com ([Bug]: interface becomes slow and unresponsive after several iterations #82))

• The /clear command fixes the problem temporarily by wiping the accumulated chat history and resetting the session state, which removes the large data structures the UI and token counter were repeatedly processing. (deepwiki.com (https://deepwiki.com/superagent-ai/grok-cli/10.1-command-reference?utm_source=openai))

So the root cause is: the CLI’s interactive UI and token-counting logic perform O(N) work over an ever-growing, unbounded chat history (including large tool outputs), and that accumulated state is only released when the user runs /clear or restarts the process.