Ai

OpenClaw v2026.4.20 Improves Setup Flow, Prompt Guidance and Session Stability

OpenClaw v2026.4.20 Improves Setup Flow, Prompt Guidance and Session Stability

OpenClaw v2026.4.20 is a focused product update that improves first-run setup, strengthens agent execution guidance, refines model cost reporting, and hardens runtime maintenance behavior. The release is especially notable for making the onboarding wizard easier to scan, preventing session-store bloat from overwhelming the gateway, and separating cron runtime state from version-controlled job definitions.

What Changed

The onboarding and setup wizard received a usability pass. The security disclaimer is now presented as a single yellow warning banner with clearer section headings and checklist-style guidance, making the most important setup advice easier to review quickly. The note body is no longer visually dimmed, a loading spinner now appears while the initial model catalog is fetched, and API key prompts include a clearer placeholder label.

Default agent guidance was also strengthened. The system prompt and the OpenAI GPT-5 overlay now place more emphasis on completion bias, checking live state before answering, recovering when early results are weak, and verifying work before finalizing a response. In practice, that should improve reliability for multi-step tasks and reduce premature or weak completions.

On the pricing side, OpenClaw now supports tiered model pricing from cached catalogs and configured models. The release also adds bundled cost estimates for Moonshot Kimi K2.6 and K2.5 in token-usage reporting, which should make cost tracking more accurate for teams using those models.

Maintenance and stability were a major focus as well. The built-in entry cap and age-based pruning are now enforced by default for session maintenance, and oversized stores are pruned during load. That change is designed to stop large cron or executor session backlogs from causing gateway out-of-memory failures before normal write-path cleanup can occur.

Elsewhere in the release, repeated same-context plugin loads now reuse plugin loader aliasing and Jiti config resolution, reducing overhead in import-heavy test scenarios. Cron runtime execution state has also been split into a separate jobs-state.json file, allowing jobs.json to remain stable for git-tracked job definitions. Finally, context compaction now supports optional start and completion notices, giving users better visibility when that process runs.

Why It Matters

This release matters because it improves both ends of the product experience: setup becomes clearer for new or returning operators, while production runtime behavior becomes safer for long-lived systems. The onboarding tweaks reduce confusion during initial configuration, especially around security warnings and API key entry.

The prompt-layer changes are equally important because they affect how agents behave during real work. By explicitly reinforcing state checks, recovery behavior, and verification before final output, OpenClaw is pushing for more dependable execution rather than faster-but-riskier responses.

The session pruning and cron-state changes are particularly relevant for operators running persistent automation. Separating mutable runtime state from tracked job definitions is cleaner operationally, and pruning oversized stores at load time addresses a failure mode that can be painful in unattended environments.

Taken together, v2026.4.20 looks like a quality and resilience release: less friction in setup, better execution discipline in agents, more accurate cost visibility, and stronger protections against runtime degradation as automation backlogs grow.

Official Source: https://github.com/openclaw/openclaw/releases/tag/v2026.4.20