Chat: Add -n/--count to show previous messages #1248
Conversation
Default to no history (maintain current behavior) Naked flag (-l,--last-messages) defaults to 5 (based on other defaults) Optional NUMBER to request different number TODO: Setting to 0 gets full history (same as logs) This code written on a Steam Deck with llm
- Defaults to 3 now, matching logs list subcommand - Passing 0 now shows full history - Each Response now shows date + id, matching logs format - BaseResponse now sets `_start_utcnow` field in `from_row` - Response's `datetime_utc` function now formats like logs - Finalize test suite This commit made on a Steam Deck w/ llm + gemini-2.5-flash
Also, -l might be used for --log to enable logging in chat
|
This falls under "Don't trust the dev to test their own code," but I've been using this feature locally since I put up the PR and have been absolutely loving it. My use case is that I made a little persona (template) to be my workout "trainer" that I check in with after each of my workouts. The system prompt is to give a little encouragement or tip. Each day I open it with (technically, I have Could I have done this via a Gemini Chat instead? Well, yeah, but why? I do everything else on the command line ;) |
2 participants
Primary Feature
Add a new option to the
chatsub-command, allowing users to view the prior prompt/responses of a resumed conversation.How to Use
When calling the
chatsub-command, simply add-n,--count(like you would for thelogssubcommand). Likelogs, it will default to the last 3 messages of the conversationAdditional Features
Count Control
You can retrieve any number of messages, or ALL of the messages using
0(like-n/--countforlogs)Safety
Calling
chatwith-n/--countwithout-c,--continue,--cid,--conversationwill simply start a new chat (since, logically, there is no history)llm chat -n 99 # 99 problems but this command ain't oneBug Fix
While formatting the output of this option, I realized that there was a bug in how
_BaseResponseentities were created from DB rows - thedatetime_utcwasn't being loaded into_start_utcnow, and the format wasn't matching the rest of the llm output. I fixed it, and included a test in my suite to validate the formatting.