OpenClaw 2026.3.8-beta.1: Safer Backups, Better Talk Mode, and Cleaner Channel Routing
OpenClaw News Editorial
OpenClaw 2026.3.8-beta.1: Beta Release Notes That Actually Reduce Pager Alerts
This is a pre-release (beta). If you run OpenClaw in production, treat this like a controlled rollout: test on a staging Gateway, snapshot config/state, then promote.
Source: GitHub release — https://github.com/openclaw/openclaw/releases/tag/v2026.3.8-beta.1
1) Backup finally becomes a first-class workflow (create + verify)
The most important operational change in this beta is new CLI support for local state archives:
openclaw backup createopenclaw backup verify
Why it matters:
- It turns “I think I backed it up” into a verifiable artifact (manifest + payload validation).
- It supports operationally sane modes like
--only-configand--no-include-workspaceso you can back up what you intend (and avoid accidentally shipping workspaces/secrets to the wrong place). - It pairs well with destructive flows (upgrade/migration) where you want a guardrail, not a hope.
Practical rollout tip:
- Before you upgrade a real Gateway, automate: create → verify → store (and make the verify step blocking).
2) Talk Mode gets a top-level silence timeout (less “why didn’t it send?”)
A new config knob lands:
talk.silenceTimeoutMs
It lets Talk wait a configurable duration of silence before auto-sending the current transcript. The key point: if unset, platforms keep their existing defaults, so you can adopt it gradually.
When this helps:
- Noisy environments where Talk “never sees silence” and transcripts get stuck.
- Team workflows where you want a predictable pause window (e.g. 1.2–1.8s) so the agent response cadence feels consistent.
3) macOS remote onboarding: token handling is less fragile
The macOS app gains a remote gateway token field for remote mode and, importantly, it preserves non-plaintext token config values until you explicitly replace them.
Translation: fewer “I opened settings and it overwrote my working secret reference” incidents.
4) Telegram routing: fewer duplicate replies in DMs
A fix changes Telegram inbound DM deduping to be per agent (not per session key), preventing a single DM from triggering duplicate replies when multiple session targets resolve.
If you’ve ever seen “the bot answered twice” in Telegram DMs, this is the patch you care about.
5) Feishu onboarding: stop re-prompting after plugin install
A targeted fix clears short-lived discovery cache before reloading the registry after installing a channel plugin.
Operationally, that reduces the annoying loop:
install plugin → onboarding says “download Feishu” again → user loses confidence
6) Web search neutrality & Brave “LLM context” mode (opt-in)
Two changes that affect how “web_search” behaves:
- Provider ordering is alphabetized to stay neutral.
- Brave search adds an opt-in config:
tools.web.search.brave.mode: "llm-context"to call Brave’s LLM Context endpoint and return grounding snippets + metadata.
If your agents rely on web grounding, this is worth testing because it can reduce hallucinations without changing prompts.
Recommended beta test checklist
- Run backup create/verify on your current Gateway and confirm restore expectations.
- Test Talk Mode with
talk.silenceTimeoutMsand measure perceived latency. - Validate your most important channel routes (Telegram DM, Feishu onboarding).
- If you use browser relay (Chrome extension), verify reconnect behavior on transient drops.
OpenClaw News Editorial
Unofficial. Practical. Ops-first.
