Release notes — OpenClaw Hybrid Memory 2026.3.150

Release date: 2026-03-15

This release adds Phase 3 modularization, OAuth-first auth with automatic failover, and clearer configuration modes. It builds on the Phase 1 core-only baseline (2026.3.140) and the Phase 2.3 staged lifecycle.

What’s new

OAuth preferred, with smart failover

When a provider has both OAuth (e.g. Claude CLI) and an API key configured, the plugin now prefers OAuth by default. If OAuth fails (gateway down, token expired, etc.), it:

Falls back to your API key so LLM calls keep working.
Records a failure and waits before trying OAuth again, using an incremental backoff: 5 min → 30 min → 1 h → 2 h → 4 h.
Resets backoff automatically after 24 hours, or you can clear it immediately with:
```
openclaw hybrid-mem reset-auth-backoff
```

You can turn off OAuth preference with auth.preferOAuthWhenBoth: false. See LLM-AND-PROVIDERS.md for auth.backoffScheduleMinutes and auth.resetBackoffAfterHours.

Clearer configuration modes

Configuration modes have new names and a single default:

New name	Best for	Notes
complete	Default — full features	Same as previous “full”; everything on.
enhanced	Most features, no docs	No MarkItDown doc ingestion.
minimal	Low cost, nano/flash only	LLM restricted to nano/flash tier.
local	No external LLM	FTS-only; zero API calls.

If you don’t set mode, the default is local (cost-safety: no external LLM).
Old names (essential, normal, expert, full) are reset to local and a one-time warning is logged. Set "mode": "minimal", "enhanced", or "complete" explicitly to enable LLM and other features.

See CONFIGURATION-MODES.md and FEATURES-AND-TIERS.md.

New and updated CLI

openclaw hybrid-mem config — Shows your effective plugin config and detected mode (or “Custom” if you’ve overridden presets).
openclaw hybrid-mem reset-auth-backoff — Clears OAuth failover state so the next LLM call tries OAuth again (for providers with both OAuth and API key).
openclaw hybrid-mem verify --test-llm — Runs a minimal completion per provider and reports OAuth result and API result separately, plus credential source (env, file, plugin, gateway, local). Respects llm.disabledProviders.

Memory-to-skills removed

The memory-to-skills feature (skills-suggest command, cron job, and related config) has been removed. Use workflow crystallization and tool proposals instead. The number of maintenance cron jobs is reduced from 9 to 8.

Under the hood

Stable internal API — Optional modules can depend on MemoryPluginAPI for tool and lifecycle registration without circular dependencies.
Staged lifecycle — Hooks are split into setup, recall, injection, capture, and cleanup stages with per-stage timeouts and toggles.
Procedure injection — Capped by procedures.maxInjectionTokens (default 500); blocks are trimmed to stay within the cap.
Fixes — Credential source fallback in verify, Google base URL for --test-llm, and removal of stale skills-suggest references.

Upgrade

Pull or install the plugin at 2026.3.150.
Modes: If you use deprecated names, you’ll see a one-time warning; set the new names when you edit config. Default remains full-featured (complete) when mode is omitted.
OAuth + API key: No change needed; OAuth is preferred and API key is used on failure with backoff. Use reset-auth-backoff if you want to retry OAuth immediately.
Memory-to-skills: Remove skills-suggest from cron and config if you used it.

Full changelog: CHANGELOG.md.