Hermes Agent v0.14.0 — The Release That Makes Everything Just Work

May 16, 2026. 808 commits, 633 merged PRs, 215 contributors, 545 issues closed. They called it "The Foundation Release," and the name fits perfectly.

I've been using Hermes Agent since the early days. v0.14.0 doesn't add flashy new features — instead, it fixes every friction point that made daily use annoying. Here's what changed.

Installing It No Longer Sucks

Before v0.14.0, getting Hermes running meant cloning a repo, running shell scripts, debugging environment issues. 30 minutes if you were lucky.

Now?

pip install hermes-agent

That's it. The wheel bundles the Ink TUI and shell launcher — no Git, no install scripts. Upgrading is pip install --upgrade hermes-agent.

Heavy dependencies (Slack, Matrix, Feishu, image gen, TTS) are now lazy-loaded — they install only when you first use them. Your base install is dramatically smaller.

Native Windows support is also here (early beta). No WSL required — it runs in cmd and PowerShell. The PowerShell installer handles MinGit, Python stub detection, and Ctrl+C signal handling automatically. Over 40 Windows-specific fixes went into this release.

hermes proxy: One Subscription, Every Tool

Claude Pro is great, but its API isn't OpenAI-compatible. That means tools like Codex CLI, Aider, Cline, and VS Code Continue couldn't use it.

hermes proxy fixes this. It spins up a local endpoint that translates OAuth providers into OpenAI-compatible API calls:

hermes proxy --provider anthropic --port 8080

Set OPENAI_BASE_URL=http://localhost:8080/v1 in any tool and you're done. Works with Claude Pro, ChatGPT Pro, and SuperGrok. One subscription, every tool.

Keep this local. Don't expose the port to the internet.

/handoff: Switch Models Mid-Conversation

Switching between AI personalities used to mean closing the session, starting fresh, losing all context.

/handoff changes that. One command migrates your entire session — messages, tool calls, context, everything — to a different model, persona, or profile:

/handoff deep-reasoning

My favorite workflow: do the grunt work with a fast model, then /handoff to a reasoning model when I hit a wall. No new window, no context lost.

Performance That Adds Up

• Cold start: ~19 seconds faster. Lazy imports, disk-cached model catalogs, parallel health checks.
• hermes tools: 14s → under 1.5s. The all-platforms check no longer makes you wait.
• Browser tools: 180x faster. A single persistent Chrome DevTools WebSocket connection replaces per-call sessions. Page interactions feel instant.

The one that surprised me: cross-session Claude prompt caching. System prompts, skills, and memory prefixes are cached for an hour across sessions. Cross-session cache hits mean /new is fast and cheap. If you use Hermes daily, the token savings add up fast.

Other Highlights

• x_search: X/Twitter search is now a first-class built-in tool. OAuth or API key, no extra skill needed.
• LSP semantic diagnostics: After every file write, a real language server runs incremental diagnostics — type errors and missing imports are caught immediately, not two turns later.
• Grok integration: SuperGrok OAuth login. grok-4.3 supports a 1M token context window.
• LINE + SimpleX Chat bring the total supported messaging platforms to 22.

Should You Upgrade?

pip install --upgrade hermes-agent
hermes --version

Back up your config.yaml first. If you're cautious, wait 24 hours for edge-case fixes.

---

v0.14.0 solves the problems you actually hit every day: can I install it, can it talk to my other tools, can I switch modes without starting over, and why is it so slow to launch. No vanity features — just the stuff that makes Hermes go from "it works" to "it's good."