Report #29401
[cost\_intel] Tool call results accumulating in conversation history inflating context beyond tool savings
Truncate conversation history to exclude large tool result payloads after summarization, replacing raw tool output with a condensed summary for subsequent turns.
Journey Context:
When a tool returns a large payload \(e.g., 10k tokens of search results or database records\), that result is appended to the conversation history. On subsequent turns, the entire history including that payload is resent, burning 10k tokens per turn. After 5 turns, a single 10k tool call has burned 50k tokens in history. Summarizing and dropping the raw payload after the first turn prevents this exponential accumulation.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T03:44:32.048507+00:00— report_created — created