Token limit exceeded because retriever output piles up in chat history

A ReAct agent with a persistent checkpointer starts failing with token-limit errors after a few turns.

Difficulty

Moderate

Time

~15 min

Steps

Fixed in

—

Last checked

2026-10-15

Unverified: we could not confirm from the sources that a fix shipped. Source: issue closed as completed, no linked fix. Last checked 2026-10-15.

Symptom

A ReAct agent with a persistent checkpointer starts failing with token-limit errors after a few turns.

Evidence what you'll see

Errors grow with conversation length, not with the question
Saved history holds large tool messages from a retriever

Cause

Every retriever result is stored as a tool message in the thread, so the context window fills up. The maintainers said this is a context-window limit, not a checkpointer bug.

Fix steps

0 of 4 done

Step 1Measure the token size of the stored messages for a failing thread.
Step 2Trim or summarize old history before each model call (the memory docs list strategies).
Step 3Limit how many documents, or how much text, the tool returns.
Step 4Optionally remove old tool calls and tool messages from chat history, as the reporter did.

Your check marks stay in this browser only.

Verify the fix worked

A long conversation that used to fail now completes.
Stored history stays under your chosen token budget.

Did this fix it?

Sources and credit

Summarized in our own words from these public reports. Credit goes to their authors.

Provided as is, without warranty. Test in a non-production environment and back up first. Security-related steps: verify against the official docs. Terms · Request removal or attribution change