Hermes v2026.4.8 Update: Hermes Agent v0.8.0 (v2026.4.8)

Release Date: April 8, 2026

This release turns Hermes from a smart assistant into a truly independent teammate — it now knows when your background jobs finish, lets you swap brain models on the fly, and even fixed its own mistakes on GPT and Codex.

🧠 What’s New (The Big Stuff)

1. Your Agent Now Taps You on the Shoulder

Remember the old way? You’d start a long training run or deployment, then sit there refreshing logs like it’s 2015. Not anymore. With notify_on_complete, Hermes can start a background task and automatically get notified when it finishes — no polling, no babysitting. Think of it like setting a kitchen timer: you go chop vegetables, and the timer tells you when the pasta’s ready. Your agent can keep working on other things and pick up results the moment they land.

2. Live Model Switching (No Restart Required)

Ever wished you could swap a car engine while driving? Now you can — sort of. The new /model command lets you switch models and providers mid-conversation from CLI, Telegram, Discord, Slack, or any gateway. It even shows interactive buttons on Telegram and Discord, so you can tap to switch. If one provider hiccups, Hermes automatically falls back to another. It’s like having a universal adapter for your AI brain.

3. Free Xiaomi MiMo v2 Pro on Nous Portal

You now get free access to Xiaomi’s MiMo v2 Pro for auxiliary tasks like compression, vision, and summarization. It’s the “helping hand” model — not the main brain, but perfect for grunt work. The model picker now shows pricing and gating clearly, so you always know what’s free.

4. Self-Optimized GPT/Codex Guidance (Yes, It Fixed Itself)

Here’s the wild part: Hermes ran automated behavioral benchmarks on GPT and Codex, found 5 failure modes in how they call tools, and patched them — without a human writing a single line of guidance. It’s like a chef tasting their own soup and adjusting the recipe. The result? Dramatically more reliable tool use on OpenAI models, plus better structured reasoning.

🔧 Improvements

Google AI Studio (Gemini) Native Provider — Direct access to Gemini models, with automatic context-length detection via the models.dev registry. No more guessing how much memory you have.
Inactivity-Based Timeouts — Old timeouts were like a parking meter: they expired even if you were actively working. Now Hermes only times out when it’s truly idle. Long-running tasks that are making progress will never be killed.
Approval Buttons on Slack & Telegram — Dangerous commands now get native platform buttons (instead of typing /approve). Slack keeps thread context; Telegram gets clean inline buttons. It’s like a physical “confirm” switch on a power tool.

🛠️ Fixes & Housekeeping

209 merged PRs and 82 resolved issues — everything from edge-case crashes to subtle race conditions.
Improved aggregator-aware routing, so Hermes stays on OpenRouter/Nous when possible, and only jumps providers when necessary.

🎯 What This Means for You

This release is about trust and autonomy. You can now delegate long tasks, walk away, and come back to results — not a dead session. You can experiment with different models without restarting your workflow. And you can rely on Hermes to be smarter about how it uses its own tools, especially on OpenAI models.

For developers, the self-optimization is a glimpse of the future: agents that debug their own behavior. For everyday users, it’s simply a more reliable, more responsive assistant.

📎 Official release notes: https://github.com/NousResearch/hermes-agent/releases