Skip to content
Cyanotype-style Sun Print (Orange Variant) style editorial illustration for the news article: Hermes Agent atomic memory and Skills Hub, OpenClaw cost

Hermes Agent atomic memory and Skills Hub, OpenClaw cost reports, and background agent tools (test in workflows)

By Harsh Desai
Share

TL;DR

From 22 to 29 June Hermes Agent added atomic batch memory edits, a redesigned Skills Hub with security scans, iMessage integration, and background subagent delegation while OpenClaw released per-agent usage-cost reporting, turn reliability fixes, and Slack relay controls.

What shipped

From 22 to 29 June both Hermes Agent and OpenClaw shipped repeated updates centered on reliability and external connections. Hermes focused on visual profile tools, atomic memory handling, and messaging bridges such as iMessage and Raft. OpenClaw emphasized mobile panels, channel overrides, and cost tracking for steadier production runs.

Hermes Agent

Nous Research delivered dozens of Hermes Agent updates across the week. The changes center on atomic memory operations that prevent edit failures, a Skills Hub with automated scans, and direct iMessage support without Mac relays. Profile building moved fully into the web dashboard while Automation Blueprints removed cron syntax for scheduling.

  • Atomic batch memory operations Hermes Agent executes multiple add, replace, and remove edits in one call against the character budget. This stops mid-edit failures and replaces fragile multi-turn updates for long agent sessions.
  • Redesigned Skills Hub browser Users explore connected hubs, featured sections, and security scans before installing skills from trusted sources like OpenAI. Full previews appear before any install step.
  • Visual profile builder in dashboard Hermes profiles for models, skills, and MCP servers are now configured without YAML files. A global switcher handles machine-wide management from one view.
  • Cursor Composer via xAI Grok The grok-composer-2.5-fast model runs inside Hermes through existing xAI OAuth. It delivers 200k context without a separate API key for coding tasks.
  • Automation Blueprints for scheduling Natural language questions replace cron syntax when setting task schedules. Blueprints render as forms in dashboard, CLI, TUI, and messaging interfaces.
  • Image-to-image edits in image_generate Existing images are modified directly by passing a source file and prompt. Background removal and color changes work across supported providers.
  • Async background subagent delegation The delegate_task tool runs subagents without blocking the main conversation. Results return as new turns once complete.
  • Desktop app with watch-windows Live subagent views, rebindable shortcuts, native notifications, and VS Code Marketplace themes are now available. The terminal pane is also resizable.
  • Raft agent network adapter Hermes joins the Raft network through a bundled platform adapter. Wake payloads carry only metadata under a privacy-by-contract design.
  • iMessage integration via Photon Hermes connects directly to iMessage using Photon's line pool. Device-code login removes the need for a Mac relay or BlueBubbles bridge.
  • Skills Hub with security scans Connected hubs and featured sections now include automated scans before skill installs from trusted providers. Previews appear before any action.
  • Web dashboard profile management Models, skills, and MCP servers are set visually without editing config files. Multiple profiles switch instantly from a unified view.
  • grok-composer-2.5-fast access The fast coding model is reached via xAI Grok OAuth inside Hermes. A reconciled 200k context window is available without extra keys.
  • Natural language task scheduling Automation Blueprints let users set schedules through questions or forms instead of cron. The same blueprints work in CLI and dashboard.
  • image_generate tool updates Source images are edited by supplying an existing file plus prompt. Changes such as background removal apply across all image providers.
  • Background subagent execution delegate_task with background=true runs long tasks while the main loop continues. Results re-enter the chat when finished.
  • Enhanced desktop workspace features Watch-windows, rebindable shortcuts, native notifications, and VS Code themes improve the desktop app. A resizable terminal pane is included.
  • Raft network gateway connection The bundled adapter links Hermes to Raft as an external agent. Metadata-only wake payloads keep message bodies private.
  • Photon Spectrum iMessage link Direct iMessage support uses Photon's managed pool and device-code auth. No Mac relay or BlueBubbles bridge is required.
  • WhatsApp Business Cloud API support Hermes connects to Meta's official hosted API. The first-party adapter removes the need for a persistent QR bridge process.
  • Upgraded memory tool batch edits Atomic add, replace, and remove operations run in one call against the budget. Fragile multi-turn memory updates are eliminated.
  • Dashboard profile configuration Complete profiles are built visually without YAML. Global switching and machine-wide management are unified in one view.
  • xAI Grok subscription model access grok-composer-2.5-fast is used directly through OAuth with a reconciled 200k window. No separate API key is needed.
  • Conversational automation blueprints Scheduled tasks are created via natural questions or native forms. Blueprints render uniformly across dashboard, CLI, and channels.
  • image_generate image-to-image edits Existing images are transformed by passing a source file and prompt. Style changes and background removal work across providers.
  • Async subagent delegation tool Background subagents are dispatched with delegate_task(background=true). The main conversation stays unblocked until results return.
  • Desktop app watch-windows upgrade Live subagent monitoring, rebindable shortcuts, native notifications, and VS Code themes are added. The terminal pane resizes.
  • Raft platform adapter integration Hermes joins the Raft network securely via the bundled adapter. Wake payloads contain only metadata under privacy rules.
  • Photon iMessage device-code login Direct iMessage connection uses Photon's pool and hermes photon login. No Mac relay or BlueBubbles bridge is required.
  • Atomic memory batch operations Multiple edits execute atomically against the character budget. Mid-edit failures are prevented in a single call.
  • Overhauled Skills Hub previews Connected hubs and featured sections offer security scans and full previews before skill installation from trusted sources.
  • Browser-based profile creation Models, skills, and MCP servers are selected visually without manual config files. Instant switching works across machines.
  • grok-composer-2.5-fast via xAI OAuth The fast coding model runs inside Hermes with a 200k context window. Existing Grok subscriptions provide access.
  • Automation Blueprints scheduling Natural questions replace cron syntax for task automation. Blueprints appear in dashboard, CLI, TUI, and messenger views.
  • image_generate source image edits Prompts modify existing images for background removal or color changes. All supported providers receive the updates.
  • Background subagent delegation Long tasks run via delegate_task(background=true) while the main loop continues. Results appear as new turns when done.
  • Advanced desktop workspace tools Watch-windows, rebindable shortcuts, native notifications, and VS Code themes enhance the app. The terminal pane resizes.
  • Raft gateway channel adapter The bundled adapter connects Hermes to the Raft network. Metadata-only payloads keep message bodies secure.
  • iMessage connection via Photon Direct iMessage support uses Photon's line pool and device-code authentication. No Mac relay or BlueBubbles bridge is needed.

OpenClaw

OpenClaw released frequent updates focused on cost visibility, turn reliability, and channel controls. Per-agent usage reporting and Slack relay mode arrived alongside Android settings panels and file-driven CLI options. Model routing and Codex approval flows also received stability fixes.

  • Per-agent usage-cost reporting OpenClaw now tracks costs per agent with channel identity hooks for precise routing and accounting in production setups.
  • Agent turn reliability upgrades Codex partial deltas and long-context prompt-cache stability reduce lost progress during extended runs.
  • Android settings detail panels New panels improve configuration visibility and control on mobile devices for on-the-go operators.
  • Externalized official plugins Additional plugins are released cleanly as standalone packages with bundled icon metadata for safer installs.
  • File-driven agent runs via CLI The --message-file option and RAFT CLI wake bridge support richer remote and scripted workflows.
  • Slack relay mode and Mattermost commands Native relay mode, /oc_queue slash command, and per-DM model overrides simplify channel automation.
  • Per-agent cost reporting hooks Usage costs are now reported per agent with channel context for accurate billing and routing.
  • Prompt-cache stability improvements Harness activation and partial deltas keep long-context runs consistent and prevent dropped progress.
  • Standalone plugin npm packages Official provider plugins are now first-class npm releases that load at gateway startup.
  • Android configuration panels Detailed settings panels give mobile users better visibility and control over agent options.
  • CLI file-driven execution The openclaw agent command accepts message files and supports RAFT wake bridges for remote starts.
  • Slack relay and Mattermost slash commands Relay mode, /oc_queue, and per-DM overrides give finer channel control without extra setup.
  • Expanded provider catalog parsing Updated resolution and encrypted reasoning support handle more live provider variants automatically.
  • Turn reliability with Codex deltas Partial deltas and cache stability reduce inconsistent runs during long agent sessions.
  • Clean plugin externalization Additional official plugins ship separately with icon metadata available to installed clients.
  • Mobile settings detail panels Android receives grouped intent-based panels that improve configuration visibility on the go.
  • File-driven CLI and RAFT bridge --message-file and wake bridge options enable scripted and remote agent paths.
  • Channel control enhancements Slack relay mode, Mattermost commands, and per-DM overrides simplify automation tuning.
  • Mobile and desktop client polish Android settings are grouped by intent, iOS notifications are cleaned, and macOS file inputs open natively.
  • CLI session status workflows Sessions can be renamed from chat, compacted explicitly, and previewed with dry-run output.
  • Rich channel delivery for Telegram Progress, reasoning, and thread output are preserved across Telegram, Discord, and Slack.
  • Codex approval flow integration SecretRefs, thread context, and typed SDK helpers make approval workflows more predictable.
  • Session state reliability fixes Pending subagent announcements and media index alignment are preserved across turns.
  • Provider plugin registry refresh Onboarding now updates registry metadata after plugin install to keep auth flows working.
  • Trusted policy preservation Hook composition retains required tool policies for approval-sensitive flows.
  • Secure channel state transitions Channel switches reset stale origin fields and keep cron awareness attached to sessions.
  • Model routing reliability updates Zai synthesis and GLM failover now follow the active catalog more consistently.
  • Automatic fast mode for short turns Short conversational turns enable fast mode automatically before returning to normal.
  • v2026.6.10 release notes Automatic fast mode for brief exchanges is now active alongside routing and state fixes.
  • Agent turn and Codex improvements Session reliability, approval flows, and channel delivery for Telegram, Discord, and Slack are strengthened.
  • Short-turn fast mode toggle The agent switches to fast mode for brief exchanges and back to normal for longer ones automatically.
  • Codex Hosted Search launch Key-free search providers remain opt-in while ClawHub skill installs keep verified provenance.
  • Session workspace rail addition Control UI gains a workspace rail plus Watch app controls and Android chat context display.
  • Standalone provider npm packages External channel plugins load at startup and StepFun is available from npm and ClawHub.
  • Telegram rich HTML delivery Messages now send rich HTML, preserve markdown, and render progress drafts safely.
  • Plugin and skill coverage expansion Zalo channel entry, Trello curl dependencies, and stale link fixes are included.
  • Mobile and desktop client updates Android settings grouping, iOS notification cleanup, and native macOS file panels ship together.
  • New CLI status workflows Sessions rename from chat, compact explicitly, and support dry-run message previews.
  • Telegram, Discord, Slack delivery Richer progress and thread output plus Slack shortcut support are now reliable.
  • Codex approval flow strengthening Thread context and typed SDK helpers make routed approvals more predictable.
  • Agent turn and session fixes Subagent announcements and media index alignment stay consistent across runs.
  • v2026.6.11-beta.2 release Incremental stability and channel fixes continue the weekly cadence.
  • v2026.6.11-beta.1 release Early beta focuses on routing and state safety improvements.
  • v2026.6.10-beta.2 release Beta testing of fast mode and cost reporting precedes the stable drop.

What this means for you

For Vibe Builders: You can now schedule tasks with natural language forms in Hermes, run background subagents without blocking chats, and connect directly to iMessage or WhatsApp. OpenClaw adds per-agent cost tracking and Slack relay mode so your automations stay visible and controllable. Test Automation Blueprints and the new Skills Hub this week to ship agent workflows faster without writing YAML or cron.

For Non-techies: Hermes now lets you manage profiles and skills from a browser dashboard and send messages on iMessage without extra hardware. OpenClaw shows exact costs per agent and adds simple Slack controls so daily automations become easier to monitor. Start with the visual profile builder or Android settings panels to reduce manual setup steps.

For Developers: Hermes added atomic memory batch edits and Raft network adapters while OpenClaw improved prompt-cache stability, Codex approval flows, and model routing. Evaluate the RAFT CLI wake bridge and per-agent cost hooks against your current stack before integrating new channels. Watch the next beta for further session state guarantees.

What to watch next

Watch for Hermes desktop theme and watch-window refinements plus OpenClaw v2026.6.11 stable. Check whether per-DM overrides and verified skill provenance reach production parity this week.

Harshs take

The week shows both projects iterating heavily on the same surfaces: memory safety, channel reach, and mobile visibility. Hermes repeats nearly identical memory and profile features across four days while OpenClaw spreads reliability fixes thinly. The second-order effect is that operators now face more surface area to test rather than deeper reliability guarantees. Builders should pick one concrete workflow, such as background delegation or cost reporting, and measure it end-to-end this week before adopting further updates.

by Harsh Desai

Sources

  • [v2026.6.11-beta.2](https://github.com/openclaw/openclaw/releases/tag/v2026.6.11-beta.2)
  • [v2026.6.11-beta.1](https://github.com/openclaw/openclaw/releases/tag/v2026.6.11-beta.1)
  • [v2026.6.10-beta.2](https://github.com/openclaw/openclaw/releases/tag/v2026.6.10-beta.2)

More AI news

Everything AI. One email.
Every Monday.

New tools. Model launches. Plugins. Repos. Tactics. The moves the sharpest builders are making right now, before everyone else.

No spam. Unsubscribe anytime.