OpenClaw Compaction Config Builder
Configure compaction checkpoints, abort-safe guards, pluggable providers, and prompt-cache integration — then export a ready-to-use YAML config. Based on OpenClaw PRs #62146, #56224, #62600, #62179, and #62227 merged April 7, 2026.
Why Use the Compaction Config Builder
Built from 5 PRs merged into OpenClaw on April 7, 2026 — covering checkpoints, abort safety, pluggable providers, prompt caching, and memory ingestion.
Pluggable Provider Registry
Choose from built-in, tiered, or custom compaction providers using the new pluggable registry introduced in PR #56224. Swap providers without changing your agent code.
Compaction Checkpoints
Configure periodic state checkpoints so agents can recover from crashes without replaying full conversation history. Added in PR #62146.
Abort-Safe Compaction
Prevent the infinite loop bug fixed in PR #62600, where cancelling tool use during compaction caused agents to retry endlessly. Configure guards, retry limits, and cooldowns.
Prompt-Cache Integration
Expose prompt-cache runtime context to context engines (PR #62179) so compaction can make smarter decisions about what to keep and what to summarize.
Session Transcript Ingestion
Feed compacted transcripts into the memory-core dreaming corpus (PR #62227) for long-term learning across sessions.
Live YAML Preview
See your complete compaction config update in real time as you adjust settings. Copy the generated YAML directly into your config.yaml.
Build Your Compaction Config
Select a provider, configure checkpoints and safety settings, then copy the generated YAML into your config. Settings are saved automatically.
Frequently Asked Questions
What is compaction in OpenClaw?
Compaction is the process of compressing an agent's conversation history to stay within context window limits. As conversations grow, older messages are summarized or pruned so the agent can continue operating without hitting token limits. The compaction system manages when and how this happens.
What are compaction checkpoints (PR #62146)?
Compaction checkpoints periodically save the compaction state to disk. If an agent crashes mid-conversation, it can resume from the last checkpoint instead of replaying the entire history. You configure the interval (how many messages between checkpoints), maximum checkpoints to retain, and how long to keep them.
What is the pluggable compaction provider registry (PR #56224)?
Previously, OpenClaw shipped a single compaction engine. PR #56224 introduced a registry that lets you swap in custom compaction providers — for example, a Redis-backed compactor for distributed deployments or a tiered compactor that keeps recent context hot. Select 'Custom' in the provider section and enter your registered provider name.
How does abort-safe compaction prevent infinite loops (PR #62600)?
PR #62600 fixed a critical bug where aborting a tool call during compaction caused the agent to enter an infinite retry loop. The abort-safety settings add a guard that detects this pattern, limits retry attempts, and enforces a cooldown between retries. It is strongly recommended to keep abort safety enabled.
How does prompt-cache context help compaction (PR #62179)?
PR #62179 exposes the prompt-cache runtime context to context engines. This means the compaction engine can see what's already cached and avoid redundantly re-summarizing content that's still in the prompt cache. This reduces latency and token usage, especially for long-running agents.
What is the recommended checkpoint interval?
For most deployments, a checkpoint interval of 30-50 messages works well. Lower intervals (10-20) are safer for mission-critical agents where crash recovery time matters, but add more I/O overhead. Higher intervals (100+) are fine for low-stakes or short-lived agents.