# Haystack v3.2.0: experimental context compaction and token budgets for Agent runs

Date: 2026-09-24

Haystack v3.2.0 was published on 2026-09-24. The release notes describe two experimental agent hooks: `SummarizationCompactor`, a compaction strategy for `CompactionHook` that replaces older turns and agent steps with LLM-written summaries while keeping the newest steps unchanged, and `TokenBudgetHook`, which stops an Agent run before the next LLM call once a token budget is spent and reports `exit_reason` as `token_budget_exceeded`. Both build on a `stop_run` state key that any hook can set. Source: [v3.2.0 release notes](https://github.com/deepset-ai/haystack/releases/tag/v3.2.0).

Relation to the directory: see [Haystack](/tools/haystack) and [Haystack vs LlamaIndex agents](/compare/haystack-vs-llamaindex-agents).

Official source: https://github.com/deepset-ai/haystack/releases/tag/v3.2.0

Related tool: [Haystack](https://multiagentguide.top/tools/haystack.md)

---

Data as of 2026-09-24. Not affiliated with listed projects. HTML version: https://multiagentguide.top/new/haystack-3-2-0