6a6410bcaa
Have the LLM use square bracketed links for internal references
2026-04-13 14:24:36 +01:00
96bb0fc403
Try and control ChatGPT's tone and "helpfulness"
2026-04-13 14:09:43 +01:00
54dab5fbdf
Use gpt 5.1 to cater for collection summary context size
2026-04-13 12:43:42 +01:00
f9448cac6f
Revert "Use smaller model so that we don't exhaust tokens"
...
This reverts commit 9c0e11bc84
2026-04-13 12:34:17 +01:00
8108645718
Shrink collection summary to fit OpenAI token limit
...
Deduplicate records by musicbrainz_id (merging formats)
and cap genres to 3 per record, reducing the summary from
~32K to ~24K estimated tokens.
2026-04-13 12:33:15 +01:00
9c0e11bc84
Use smaller model so that we don't exhaust tokens
2026-04-13 12:06:24 +01:00
34f484cc81
Add CollectionChat StreamProvider
2026-04-13 10:09:50 +01:00
55be2a4bab
Add :collection to chat entity enum
2026-04-13 10:04:37 +01:00
4be9db6417
Refactor DefaultPrompt into Chats.Prompt with build/2
2026-04-05 08:32:03 +01:00
f9155f8c52
Try and control the chat responses style
2026-04-04 23:35:46 +01:00
ec12be613b
Extract a default prompt
2026-04-04 23:35:30 +01:00
86fd3134f1
Reduce repetitive LLM chat responses ( #146 )
...
Replace "always provide a link" with conditional linking and add
instructions to vary response style and avoid restating information
already covered in the conversation.
Co-authored-by: Claude <noreply@anthropic.com >
2026-04-04 18:32:57 +01:00
7d55dfd9ae
Have the AI always provide at least one link when answering questions
2026-03-19 14:30:04 +00:00
63c95ff1e2
Move all chat related code under chats
2026-03-17 14:30:24 +00:00
3f95aee7a7
Cap message content
2026-03-17 13:51:38 +00:00
8fd8dde73d
Persist chat history
2026-03-17 13:50:01 +00:00
Claudio Ortolina