OpenClaw 2026.3.8
ACP Provenance, Backup Tools & 30+ Fixes
Major ACP enhancement with metadata tracking
Release Highlights
🔍
ACP Provenance
Track session origins & lineage
💾
Backup CLI
Create & verify archives
🔎
Brave LLM Context
Grounding snippets mode
30+
Platform Fixes
macOS, Android, Telegram
ACP Provenance — What Is It?
Track where ACP sessions come from and preserve context
🔗
Session Origins
Track which agent/channel spawned each ACP session. Full lineage preservation across agent boundaries.
🧵
Trace IDs
Every ACP run gets a unique session trace ID for debugging multi-agent workflows.
📋
Metadata Injection
Optional: inject origin context into ACP agent prompts for full transparency.
👁️
Receipt Mode
Make provenance visible to the ACP agent itself — full audit trail in transcripts.
ACP Provenance — How It Works
Three modes: off, metadata, or full receipt
# Disable provenance (default for now)
openclaw acp --provenance off
# Track origin silently (metadata only)
openclaw acp --provenance meta
# Inject visible context into agent prompt
openclaw acp --provenance meta+receipt🔕
Off Mode
No tracking. ACP runs as before with no provenance metadata.
📊
Metadata Mode
Origin tracked in session state. Lineage preserved. Agent doesn't see it.
📝
Receipt Mode
Full transparency: origin context injected into agent prompt. Visible in transcripts.
💾
Persistence
Transcript archives include provenance data for audit trails.
ACP Provenance — Use Cases
Why you need this
🐛
Debug Multi-Agent Workflows
Trace errors back to the originating agent/channel. No more "where did this come from?"
📜
Audit ACP Chains
Full session lineage for compliance. Know exactly who spawned what, when.
🔍
Root Cause Analysis
When ACP sessions fail or produce bad output, trace back to the source instantly.
🧠
Context Retention
Preserve requester context across agent boundaries. ACP agents know their origin.
📊
Analytics & Logging
Track ACP usage patterns. Which agents spawn the most sessions? Where do failures cluster?
🔐
Security & Compliance
Audit trail for sensitive operations. Know who authorized each ACP action.
ACP Provenance — Benefits
Production-grade multi-agent orchestration
🔍
Better Debugging
Trace IDs link sessions to origins. No more mystery failures.
👁️
Transparency
Optional receipt mode shows agents their own lineage.
📜
Audit Trail
Full session history preserved in transcripts.
🧠
Context Preservation
Requester context flows across agent boundaries.
⚙️
Opt-In Control
Three modes: off, metadata-only, or full receipt.
🚀
Production Ready
Designed for complex multi-agent pipelines at scale.
Backup System
New CLI commands for config & workspace archives
# Create full backup (config + workspace)
openclaw backup create
# Config-only backup (fast)
openclaw backup create --only-config
# Exclude workspace (config + state only)
openclaw backup create --no-include-workspace
# Verify backup integrity
openclaw backup verify backup-2026-03-08.tar.gz📦
Date-Sorted Archives
Backups named with ISO timestamps. Easy to identify and sort chronologically.
✅
Manifest Validation
Verify command checks manifest integrity and payload completeness before restore.
⚠️
Destructive Flow Guidance
Restore operations warn about overwrites. Verification required before proceeding.
Backup Features
Production-grade backup & restore
📅
Date-Sorted Naming
backup-YYYY-MM-DD-HHMMSS.tar.gz format
⚙️
Config Snapshot
Preserves ~/.openclaw config state
📂
Workspace Archives
Optional workspace inclusion (can be large)
✅
Manifest Validation
Verify integrity before restore
🔒
Local State
Captures session state & plugin data
⚠️
Destructive Warnings
Restore requires explicit verification
⚡
Config-Only Mode
Fast backups without workspace data
🎯
Selective Exclusion
--no-include-workspace flag available
Web Search: Brave LLM Context
New mode for grounding AI responses
🧠
LLM-Optimized Results
Calls Brave's LLM Context endpoint. Returns grounding snippets instead of raw search results.
📚
Source Metadata
Includes source URLs, titles, and context. Better attribution for AI-generated content.
⚙️
Opt-In Feature
Enable in config. Existing search behavior unchanged. Choose what works for your workflow.
tools:
web:
search:
brave:
mode: "llm-context" # or "standard"
Better grounding for AI responses. Brave's LLM Context mode is purpose-built for this use case.
Platform Fixes: macOS
Stability and reliability improvements
Gateway Auth
Remote token field added
LaunchAgent
Restart recovery improved
Browser Proxy
Routing fixes applied
- Permission state refresh — better handling of macOS security prompts
- Tailscale Serve improvements — more reliable remote access
- Browser CDP WebSocket normalization — consistent connection handling
- Extension relay improvements — Chrome tab reconnect handling
macOS users get smoother gateway operations, better browser integration, and more reliable background service management.
Platform Fixes: Android
Play Store compliance and performance
Self-Update
Removed for Play Store compliance
Foreground Service
Narrowed scope for efficiency
Permissions
Location permissions cleaned up
- Background capture removed — better battery life and privacy compliance
- Deferred foreground-service startup — faster app launch times
- WebView debug init moved off critical path — improved performance
- Startup macrobenchmark + low-noise perf CLI — better diagnostics
Android build is now fully Play Store compliant with better performance and cleaner permission model.
Platform Fixes: Messaging
Better reliability across all channels
Telegram
DM deduplication fixed
Matrix
DM routing corrected
Mattermost
Thread replies working
Feishu
Plugin onboarding improved
Cron
Announce delivery fixed
Discord
Reusable buttons enhanced
Messaging reliability improved across the board. No more duplicate DMs, broken threads, or missed cron announcements.
Browser Tools
CDP improvements and WSL2 support
🔌
CDP WebSocket
WebSocket normalization for consistent Chrome DevTools Protocol connections across platforms.
🐧
WSL2 Support
Bind address support for WSL2 environments. Run OpenClaw in Linux, control Windows browsers.
🔄
Extension Relay
Improved Chrome extension relay. Better tab reconnect handling and state management.
🐳
Browserless
Container fixes for headless browser automation. More reliable remote browser control.
Browser automation is more reliable across all deployment scenarios — local, remote, containerized, and WSL2.
Developer Experience
Small improvements that add up
TUI Theme
Light theme detection added
Workspace
Agent workspace inference improved
Talk Mode
Silence timeout configurable
- Version output now includes git hash — easier debugging and issue reporting
- Context engine plugin bootstrap — faster startup for custom plugins
- Better error messages across CLI commands — clearer guidance when things go wrong
- Improved logging for debugging multi-agent workflows
The little things matter. Better defaults, clearer output, and smarter inference make daily development smoother.
Breaking Changes & Notes
Smooth upgrade path
No breaking changes! This release is pure improvements. All changes are backward compatible, all new features are opt-in, and existing configurations are preserved.
- ✅ ACP provenance is opt-in — defaults to off, enable with
--provenanceflag - ✅ Brave LLM Context mode is opt-in — existing search behavior unchanged
- ✅ Backup commands are new additions — no impact on existing workflows
- ✅ Platform fixes are transparent — no config changes required
- ✅ All existing scripts, integrations, and automations continue to work
Upgrade with confidence. Test the new features when you're ready. No rush, no risk.
Getting Started
Upgrade and explore the new features
OpenClaw 2026.3.8 is ready. Update now and try the new capabilities.
# Update to latest version
openclaw update
# Try ACP provenance tracking
openclaw acp --provenance meta+receipt "your task"
# Create a backup
openclaw backup create
# Verify a backup before restore
openclaw backup verify backup-2026-03-08.tar.gz# Enable Brave LLM Context mode in config
tools:
web:
search:
brave:
mode: "llm-context"
📚 Full documentation: docs.openclaw.ai
💬 Community: Discord, GitHub Discussions
🐛 Issues: github.com/openclaw/openclaw