Hermes v0.8 DESTROYS OpenClaw?

Hermes agent just dropped version 0.8 and this is one of the biggest updates they've shipped. 209 merged pull requests, 82 resolved issues and a handful of features that are going to genuinely change how you use AI agents day to day. So let me walk you through what actually matters here because there's a lot to unpack. The headline feature is background process auto notifications.

They're calling it notify on complete. Here's what that means in plain English. You kick off a long task, maybe you're running a report process in a batch of leads or doing a content audit and instead of the agent just sitting there waiting or you having to check back in, it now gets a notification the moment that task finishes. The agent keeps working on other things and when the job's done it picks up the results automatically.

So no polling, no babysitting, just set it and come back to the output. It sounds small but it's actually huge because right now most AI agents either block everything whilst they wait or they silently fail and you don't even know it. This fixes that. The second big one is live model switching.

So let's say you're at mid-session, you know CLI or Telegram or Discord or Slack and you type forward slash model. Instantly you can switch providers. You can switch between OpenRouter, Noose, Google, XAI, Telegram and Discord even get interactive buttons so you can tap and switch without typing anything and it's aggregator aware so it tries to stay on your preferred provider first before falling back. For anyone running Hermes across multiple platforms or testing different models for different tasks, this is a genuine time saver.

Now here's one people are going to sleep on. Self-optimized GPT and codex tool calling. The team ran automated behavioral benchmarking. Essentially the agent tested itself and found five specific failure models in how it calls tools on OpenAI models and then it patched them.

The reliability improvement on GPT and codex models is real. If you've been running Hermes with OpenAI as your back end and noticed it occasionally dropping the ball on tool use, this update is directly for you. If you want to get hands-on with Hermes right now with step-by-step coaching calls, a community of 2,800 business owners already using agents like this for lead gen content and client work, go to the AIProfitBoardroom.com or hit the link in the comments and description. We've got a 30-day roadmap specifically built around running AI agents like Hermes in your business and plenty of members already deep into Hermes and OpenCLR.

Google AI Studio is now a native provider. Gemini models, direct access, automatic context length detection via models.dev, no workarounds, just add your Google AI Studio and you're live. For anyone who's been wanting to run Gemini as a brain inside Hermes without jumping through hoops, it's now a first-class option. Inactivity-based timeouts just got smarter too.

So old behavior was that the agent would timeout based on wall clock time, meaning if you had a long-running task, something building a report or crawling a site, processing files for example, it could get killed even though it's actively working. The new behavior is a timeouts only trigger when the agent is genuinely idle. If it's doing something, it keeps going. If it's just sitting there doing nothing, it stops.

Way less frustrating. The approval flow on Slack and Telegram also got a real upgrade. So instead of having to type forward slash approve every time the agent wants to run something that needs sign off, you now get native buttons which is a lot easier because you can just tap it and you're done. Slack even preserves your thread context so the conversation stays clean.

And here's something worth paying attention to if you use MCP servers. Hermes now supports MCP OAuth 2.1 with PKCE, full standards compliant authentication for MCP server connections. And they added automatic malware scanning for MCP extension packages via the OSV vulnerability database. As the MCP ecosystem grows and more people are installing third-party MCP tools, this matters.

You're not just trusting a package is clean. Hermes is actually checking. The memory system got a new option too. SuperMemory is now a supported memory provider.

Multi-container support search mode per user identity templates. If you're running Hermes for multiple clients for example or multiple workflows and you need memory to stay properly isolated, this is the thing to look at. Execute code on remote backends is live so you can now run code execution on Docker, SSH, Modal and other remote backends. Not just locally.

And for anyone running Hermes in a cloud environment or offloading compute to a remote machine, this opens up a lot. Firecrawl is now a cloud browser provider option. And they've switched the managed provider for browsers from browser base to browser use. So if you're using Hermes for web research for example or any sort of research or browser automation, the underlying infrastructure just got a big upgrade.

There's also a new pre-run script injection for cron jobs. This lets you collect data or detect changes before the main cron task runs. So you can build workflows where the agent checks if something has actually changed before deciding whether to execute. Smart automation but fewer unnecessary runs essentially.

Matrix now has tier one feature parity. So reactions, read receipts, rich formatting, room management. Signal gets full media delivery and Mattermost gets file attachments. So whatever messaging platform you're running through Hermes, the release tightened everything up.

Centralized logging landed also. So all logs now go to a single location. Agent.log and errors.log with a Hermes logs command. So you can actually tail and filter them.

And they added config structure. So if your YAML is malformed, you get a clear error message instead of a cryptic failure later. And the plugin system expanded too significantly. So plugins can now register their own subcommands for CLI.

They get request scoped API hooks with correlation IDs. They can prompt for required environment varimentals during install. And they hook into session lifecycle events. If you're building on top of Hermes or using community plugins, this makes the whole ecosystem more capable.

One more thing worth calling out. The team ran a full documentation audit. Fixed over 40 discrepancies between the docs and the actual code base. Rewrote the guides section.

Added three new tutorials. So if you've ever gone through the Hermes docs and found something that doesn't quite match how the tool actually behaved, a lot of that is cleaned up now. To update, just run Hermes update inside your terminal and that's pretty much it. But here's the bigger picture here.

So Hermes is releasing at a pace that's genuinely hard to keep up with. Every few days, another drop. The gap between what an AI agent could do six months ago and what Hermes can actually do today, background notifications, live model switching, smart timeouts, OAuth compliant MCP execution, is not a small gap. And the businesses that figure out how to wire these tools into the actual workflows right now are building real advantages, not theoretical ones.

The agent handles background tasks. You handle the decisions. And if you want to learn exactly how to do that, how to set up Hermes, how to use a new background feature and the notification system, how to build cron workflows that actually run whilst you're not watching, come join us in the AI Profit Boardroom. We've got members actively building with Hermes right now, using it for lead generation, client deliverables, getting customers, etc.

and content pipelines. You get four weekly coaching calls every week where you can get your specific setup reviewed. You get daily tutorials walking you through the newest features as they drop. And a 30-day roadmap to get your first AI agent workflow running in your business.

And a member map so you can connect with other Hermes users near you. Link in the comment subscription or go to the airprofitboardroom.com. Hermes version 0.8 is out. 209 PRs.

The agent just got a lot smarter.