Just 9 days after the massive Tenacity release, Nous Research has shipped Hermes Agent v0.14.0 — the “Foundation Release” — and the name is deserved. This isn’t a single-feature release; it’s a sweeping infrastructure reset that makes Hermes Agent installable, runnable, and connectable from almost anywhere.
The numbers alone are staggering: 808 commits, 633 merged PRs, 1,393 files changed, 165,061 insertions, 545 issues closed (12 P0, 50 P1), 215 community contributors — all in 9 days since the Tenacity release. The project now sits at 155,609 GitHub stars (+17,000 since Tenacity), with 24,980 forks and 11,989 open issues reflecting a rapidly scaling community — growth that mirrors the wider ecosystem.
Here is the full breakdown of what the Foundation Release delivers.
🤖 xAI Grok via SuperGrok OAuth — With a 1M Context Window
The headliner. If you pay for SuperGrok, you can now use xAI Grok inside Hermes Agent — positioning the project among the top open-source AI agent frameworks — by signing in with your xAI account — no API key, no separate billing. The integration goes through OAuth, so your subscription credentials are handled securely without exposing raw tokens.
But the real surprise: Grok’s context window jumps to 1M tokens through this wire-through. That means you can drop entire codebases or research corpora into a single Grok prompt — a capability that puts Grok on par with Gemini 2.5 Pro for long-context use cases. (#26534, #26664, #26644)
“Proper handling for entitlement errors and an SSH-to-tunnel docs page for when you’re SSH’d into a remote box and need to complete the OAuth flow.” — Release notes
For remote-server users, the OAuth flow also includes an SSH tunneling guide so you can authenticate from a headless machine through your local browser.
🏪 OpenAI-Compatible Local Proxy: One Subscription, Every Tool
PR #25969 introduces hermes proxy — a local HTTP server that speaks the OpenAI API format but is backed by whichever OAuth provider you’re signed into. This is deceptively powerful:
| Subscription | Becomes available to | Without |
|---|---|---|
| Claude Pro | Codex CLI, Aider, Cline, Continue | An Anthropic API key |
| ChatGPT Pro | Any OpenAI-compatible tool | An OpenAI API key |
| SuperGrok | Any OpenAI-compatible tool | An xAI API key |
Run hermes proxy, get a http://localhost:port endpoint, and point any tool that expects an OpenAI-compatible API at it. One subscription, every tool — no additional API keys required. This effectively turns your Hermes Agent subscription into a universal API gateway for the entire agent tool ecosystem.
🐦 x_search: First-Class X (Twitter) Search
The agent can now search X/Twitter directly — no skill installation, no custom integration wiring. PR #26763 adds x_search as a built-in gated tool with two auth modes:
- X OAuth login — use your existing Twitter session
- API key — traditional bearer token auth
Search the timeline, find threads, surface specific posts — all from the chat interface. For journalists, researchers, and social media analysts running Hermes Agent as their primary interface, this eliminates an entire category of manual workflow.
🏢 Microsoft Teams: End-to-End
Hermes Agent now speaks Microsoft Teams natively. This isn’t a partial integration — the full stack lands together:
- Microsoft Graph auth + client foundation (#21922)
- Webhook listener that receives Teams events (#21969)
- Pipeline plugin runtime for processing Teams messages (#22007)
- Outbound delivery to post back (#22024)
Wire up the bot once, then chat to your agent from any Teams channel, DM, or group. This is a massive step for enterprise adoption — Microsoft Teams is the corporate standard, and now Hermes Agent sits inside it as a first-class citizen.
📦 PyPI Packaging: pip install hermes-agent && hermes
One of the most requested features has landed: Hermes Agent is now a real PyPI package. No more cloning the repo or running shell installers. One pip install hermes-agent command and you’re running. The wheel ships with the Ink TUI bundle and shell launcher, so the full interactive experience comes out of the box. (#26593)
This is paired with a comprehensive debloating wave (#24220, #24515):
- Heavy backends (Slack/Matrix/Feishu SDKs, image-gen, voice/TTS providers) now lazy-install on first use
- The
[all]extras drop everything covered by lazy-deps - Tiered install fallback when a wheel doesn’t fit your platform
- Supply-chain advisory checker scans every install for unsafe versions
The net effect: faster installs, a smaller disk footprint, fewer transitive vulnerabilities, and a dramatically lower barrier to entry.
⚡ Performance: Cache, Cold Start, and Browser
Cross-Session 1-Hour Claude Prompt Cache
PR #23828 brings cross-session prompt caching to Claude (via Anthropic, OpenRouter, or Nous Portal). The prompt prefix (system prompt, skills, memory) now caches for one full hour across sessions. Start a /new session and the first response comes back faster and cheaper because the cache is still warm. Background memory review hits the cache too — no full-price turns. (#25434)
Cold-Start Wave: ~19 Seconds Off hermes Launch
A coordinated performance pass across the launch path:
| Optimization | Improvement |
|---|---|
| Heavy adapters deferred to first use | Major import-time reduction |
| Model catalogs from disk cache first | Eliminates network wait |
| Doctor checks run in parallel | Concurrent connectivity probes |
chat -q skips welcome banner |
Instant prompt |
hermes tools All-Platforms screen |
14s → under 1.5s |
180× Faster Browser Console Evaluations
PR #23226 routes browser console evaluations through the supervisor’s persistent CDP WebSocket instead of spinning up a new DevTools session per call. Things that used to take seconds now return in milliseconds — real-world page interactions feel instantaneous.
💬 Two New Platforms: LINE + SimpleX Chat
Hermes Agent adds two more messaging platforms, bringing the total to 22:
- LINE (#23197) — huge in Japan, Korea, and Taiwan. Native integration with the LINE Messaging API.
- SimpleX Chat (#26232) — the privacy-focused decentralized messenger with no user IDs, no servers, no directory.
Both are full first-class platforms with inbound + outbound messaging.
🪟 Native Windows Beta
PR #21561 brings Hermes Agent to native Windows — running on cmd.exe and PowerShell without WSL. A full PowerShell installer handles MinGit auto-install, Microsoft Store Python stub detection, and Ctrl+C handling. There are still rough edges (this is early beta), but the basic loop works end-to-end on a clean Windows box.
🧠 Smarter Agent: LSP Diagnostics, Handoff, and File Verification
Three improvements that make the agent dramatically more reliable:
| Feature | What it does | PR |
|---|---|---|
| LSP Semantic Diagnostics | After every write_file/patch, runs a real language server against the file and surfaces errors (type errors, undefined symbols, missing imports) |
#24168 |
/handoff Live Session Transfer |
Moves the active session — every message, tool call, and context — to a different model/persona/profile mid-conversation | #23395 |
| Per-Turn File-Mutation Verifier | After every turn that wrote/edited files, a footer summarizes exactly what changed on disk — paths, line counts, deltas | #24498 |
The LSP diagnostics are a particularly consequential improvement. v0.13.0 had basic Python/JSON/YAML/TOML linting; v0.14.0 upgrades to actual semantic analysis — the same kind of error detection your IDE gives you, now inside the agent’s loop.
🖥️ Computer Use for Non-Anthropic Models
The computer_use tool (agent-controlled mouse + keyboard for GUI apps) used to be locked to Anthropic’s SDK. The new cua-driver backend (#21967) works with non-Anthropic providers too, with proper focus-safe operations and auto-refresh on hermes update. Now any vision-capable model can drive your desktop.
🎥 New Capabilities: Video Generation, Vision Pixels, and Web Search
A trio of new capabilities rounds out the Foundation Release:
video_generatewith pluggable backends (#25126) — one tool, any video model, with plugin architecture for future providersvision_analyzereturns raw pixels (#22955) — vision-capable models (GPT-5, Claude, Gemini, Grok-vision) now get actual pixels instead of text-summary round-trips- Brave Search (free tier) + DuckDuckGo (#21337) — two new free web-search backends join Tavily, SearXNG, and Exa
🔒 Security: Sudo Brute-Force Block, Tool Error Sanitization, and More
Three significant security improvements:
- Sudo brute-force block — the approval gate now blocks
sudo -Sbrute-force attempts and classifies stdin-fed/askpass-stripped sudo invocations as DANGEROUS (#23736) - Three dangerous-command bypasses closed — inspired by Claude Code’s command-detection work (#26829)
- Tool error sanitization — error strings are sanitized before being re-injected into the model context, preventing malicious files or remote services from passing instructions through error output (#26823)
📊 Growth Summary (May 7 → May 18)
| Metric | May 7 (v0.13.0) | May 18 (v0.14.0) | Change |
|---|---|---|---|
| GitHub Stars | ~138,000 | 155,609 | +17,609 |
| Forks | ~22,000 | 24,980 | +2,980 |
| Issues Closed | — | 545 (12 P0, 50 P1) | — |
| Merged PRs | — | 633 | — |
| Contributors | — | 215 | — |
| Messaging Platforms | 19 | 22 | +3 |
| Inference Providers | ~20 | ~25 | +NovitaAI, Grok OAuth |
🔭 What’s Next
The Foundation Release name is apt: v0.14.0 builds the infrastructure that the next wave of features will sit on top of. With PyPI packaging, a local OpenAI-compatible proxy, native Windows support, and 22 messaging platforms, Hermes Agent has the broadest surface area of any open-source agent runtime.
Areas to watch for v0.15.0:
- Frontdesk runtime — the always-available background worker architecture (still in PR)
- Kanban maturity — suspend/resume primitives for long-running task lifecycle
- Cron expansion — per-job reasoning, name-based lookup, conditional gates
- More providers — the NovitaAI addition suggests growing the long-tail of provider support
- 160K stars — at current growth (~1,700/day), this milestone falls within days
For developers building agentic systems: the Foundation Release is the release you’ve been waiting for to integrate Hermes Agent into your stack — it installs cleanly, connects everywhere, and runs on every major operating system.
Coverage based on the Hermes Agent v0.14.0 (v2026.5.16) release. Star counts as of 2026-05-18 09:00 UTC. Full changelog: v2026.5.7…v2026.5.16