BYOK Conversation Summary Failing on Letta Cloud #3111
Description
Describe the bug
When using Letta Cloud with a Bring Your Own Key (BYOK) OpenAI configuration, conversation summary does not run or trigger. The same agent and workflow successfully runs conversation summary when switched to a Letta-provided model via the Cloud model dropdown. This appears to be specific to BYOK-backed models rather than the model capability itself.
Please describe your setup
-
How are you running Letta?
- Desktop (Letta Cloud)
-
Describe your setup
- OS: Windows 10/11
- Environment: Letta Cloud (web UI, not self-hosted)
- No Docker or local runtime involved
Screenshots
N/A (behavioral issue; can provide UI screenshots if requested)
Additional context
-
Model tested:
- OpenAI GPT-5.2 via Letta Cloud dropdown → conversation summary works
- OpenAI GPT-5.2 via BYOK → conversation summary does not run
-
This suggests the issue is not with GPT-5.2 itself, but with how BYOK models are invoked for conversation summarization in Letta Cloud.
-
No agent configuration changes were made between tests other than switching the model source (Letta-provided vs BYOK).
Agent File (optional)
Not attached yet, but can be provided on request. The issue reproduces across agents.
If you're not using OpenAI, please provide additional information on your local LLM setup:
Local LLM details
N/A — this issue occurs on Letta Cloud only and does not involve local LLMs.
My read: this smells like a Cloud-side routing or permission issue where the conversation summary job is not being dispatched to BYOK-backed models, even though they’re functionally identical at inference time. I’d treat it as a backend integration bug rather than user misconfiguration.