Fimeg/letta-server
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

New prompts/defaults for sliding_window and all compaction (#9444)

Browse Source 
* new prompts for sliding window and all compaction + defaults to corresponding prompt

* regenerate api spec

---------

Co-authored-by: Amy Guan <amy@letta.com>
This commit is contained in:
amysguan
2026-02-12 18:29:30 -08:00
committed by Caren Thomas
parent 0b08164cc2
commit 9bec8c64f5
3 changed files with 71 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
fern/openapi.json
View File

@@ -30108,10 +30108,16 @@
"description": "Optional model settings used to override defaults for the summarizer model."
},
"prompt": {
"type": "string",
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"title": "Prompt",
"description": "The prompt to use for summarization.",
"default": "The following messages are being evicted from your context window. Write a detailed summary that captures what happened in these messages.\n\nThis summary will appear BEFORE the remaining recent messages in context, providing background for what comes after. Include:\n\n1. **What happened**: The conversations, tasks, and exchanges that took place. What did the user ask for? What did you do? How did things progress?\n\n2. **High level goals**: If there is an existing summary in the transcript, make sure to take it into consideration to continue tracking the higher level goals and long-term progress. Make sure to not lose track of higher level goals or the ongoing task.\n\n3. **Important details**: Specific names, data, configurations, or facts that were discussed. Don't omit details that might be referenced later.\n\n4. **Lookup hints**: For any detailed content (long lists, extensive data, specific conversations) that couldn't fit in the summary, note the topic and key terms that could be used to find it in message history later.\n\nWrite in first person as a factual record of what occurred. Be thorough and detailed - the goal is to preserve enough context that the recent messages make sense and important information isn't lost.\n\nKeep your summary under 250 words. Only output the summary."
"description": "The prompt to use for summarization. If None, uses mode-specific default."
},
"prompt_acknowledgement": {
"type": "boolean",
@@ -30228,10 +30234,16 @@
"description": "Optional model settings used to override defaults for the summarizer model."
},
"prompt": {
"type": "string",
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"title": "Prompt",
"description": "The prompt to use for summarization.",
"default": "The following messages are being evicted from your context window. Write a detailed summary that captures what happened in these messages.\n\nThis summary will appear BEFORE the remaining recent messages in context, providing background for what comes after. Include:\n\n1. **What happened**: The conversations, tasks, and exchanges that took place. What did the user ask for? What did you do? How did things progress?\n\n2. **High level goals**: If there is an existing summary in the transcript, make sure to take it into consideration to continue tracking the higher level goals and long-term progress. Make sure to not lose track of higher level goals or the ongoing task.\n\n3. **Important details**: Specific names, data, configurations, or facts that were discussed. Don't omit details that might be referenced later.\n\n4. **Lookup hints**: For any detailed content (long lists, extensive data, specific conversations) that couldn't fit in the summary, note the topic and key terms that could be used to find it in message history later.\n\nWrite in first person as a factual record of what occurred. Be thorough and detailed - the goal is to preserve enough context that the recent messages make sense and important information isn't lost.\n\nKeep your summary under 250 words. Only output the summary."
"description": "The prompt to use for summarization. If None, uses mode-specific default."
},
"prompt_acknowledgement": {
"type": "boolean",