v0.14.0

Hermes Agent v0.14.0 — The Foundation Release

Overview

v0.14.0 — The Foundation Release. Released May 16, 2026. 808 commits · 633 merged PRs · 1,393 files changed · 165,061 insertions · 545 issues closed (12 P0, 50 P1) · 215 community contributors.

Hermes Agent v0.14.0 isn’t about flashy features — it’s about making the entire experience feel solid. The theme is straightforward: Hermes installs and runs anywhere, ships with the things you actually want, and stops shipping the things you don’t. pip install hermes-agent works from PyPI for the first time. Heavy backends lazy-install on first use. The cold-start path shaves ~19 seconds off launch. Browser CDP calls are 180x faster. And the foundation for everything else — providers, messaging platforms, tools, plugins, skills — gets a serious upgrade.

Dubbed “The Foundation Release” by the team, v0.14.0 answers the question: can an agent framework graduate from “cool project” to “something you depend on every day”?

Major Features

1. PyPI Installation & Debloating Wave

Hermes Agent is now a real PyPI package. One command and you’re running:

pip install hermes-agent
hermes

The wheel ships with the Ink TUI bundle and shell launcher out of the box — no cloning repos, no shell installers. Upgrading is pip install --upgrade hermes-agent.

More importantly, the base install is dramatically lighter. Heavyweight backends (Slack, Matrix, Feishu, DingTalk adapters, image-gen SDKs, voice/TTS providers, Camofox, Codex app-server) now install automatically the first time you actually use them. The [all] extras drop everything covered by lazy-deps, and the installer falls back through tiers when a wheel doesn’t fit your platform. A supply-chain advisory checker scans every install for unsafe versions.

What this means in practice: faster installs, smaller disk footprint, fewer transitive vulnerabilities, and a dramatically lower barrier to entry for new users. The days of “I’ll try it later, the install is too heavy” are over.

Related PRs: #24220, #24515, #25014, #25038, #25766, #21818, #26593, #26148.

2. xAI Grok via SuperGrok OAuth — 1M Context Window

If you pay for SuperGrok, you can now use Grok inside Hermes by signing in with your xAI account — no API key, no separate billing. The wire-through also bumps grok-4.3 to a 1M token context window, so you can drop whole codebases or research corpora into a single prompt.

# After signing in via OAuth, just pick Grok from the model selector
hermes model  # select xAI > grok-4.3

Includes 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.

Related PRs: #26534, #26664, #26644, #26592.

3. OpenAI-Compatible Local Proxy for OAuth Providers

The headline feature for multi-tool workflows: hermes proxy spins up a local endpoint that speaks the OpenAI API but is backed by whichever OAuth provider you’re signed into — Claude Pro, ChatGPT Pro, SuperGrok.

# Start the proxy
hermes proxy --provider anthropic --port 8080

# Now any OpenAI-compatible tool just works
export OPENAI_BASE_URL=http://localhost:8080/v1

Any tool that expects an OpenAI-compatible endpoint (Codex CLI, Aider, Cline, Continue, your custom scripts) can now use your existing subscription. One subscription, every tool. This is a big deal for developers who switch between coding assistants — no more juggling API keys per tool, no more separate billing. Hermes becomes the unified routing layer between your subscriptions and your workflow.

Keep the port local — don’t expose it to the public internet.

Related PR: #25969.

4. x_search — First-Class X (Twitter) Search

The agent can now search X directly without installing a skill or wiring up a custom integration. Search the timeline, find threads, surface specific posts — straight from the chat.

# In conversation with the agent
"Search X for what people are saying about the new Claude release"

Auth works with either your X OAuth login or an API key, whichever you have. This graduates from “community hack” to first-class built-in tool.

Related PR: #26763.

5. Microsoft Teams — End-to-End

The full Microsoft Graph stack lands together: auth + client foundation, a webhook listener that receives Teams events, a pipeline plugin runtime, and outbound delivery. Wire up the bot once, then chat to your agent from any Teams channel, DM, or group.

This completes the Teams integration that had been partially available since v0.12.0 — the foundational pieces are now all in place for production use.

Related PRs: #21922, #21969, #22007, #22024.

6. Performance: ~19s Faster Cold Start, 180x Faster Browser

Two major performance threads in this release:

Cold start optimization (~19 seconds saved):

  • Heavy adapter imports deferred to first use (Feishu, QQ, Yuanbao, Google Chat, Teams, fal_client)
  • Model catalogs served from disk cache first, skipping network when fresh
  • Doctor connectivity checks run in parallel with IMDS disabled
  • chat -q single-query mode skips the welcome banner entirely
  • hermes tools All-Platforms screen: 14s → under 1.5s

Browser CDP: 180x faster:

All browser_console evaluations now share one persistent WebSocket connection to Chrome instead of spinning up a new DevTools session every call. Page interactions that used to take seconds return in milliseconds.

Related PRs: #22138, #22120, #22681, #22790, #22808, #22831, #22859, #22904, #22766, #25341, #23226.

7. Cross-Session 1-Hour Claude Prompt Cache

When you use Claude through Anthropic, OpenRouter, or Nous Portal, the prompt prefix (system prompt, skills, memory) now caches for an hour across sessions. Start a /new session and the first response comes back faster and cheaper because the cache is still warm from your last session. Background memory review hits the cache too.

If you use Hermes with Claude daily, the token savings add up fast — every /new session reuses the expensive system prompt prefix instead of paying full price.

Related PRs: #23828, #25434, #24778.

8. LINE + SimpleX Chat — 22 Messaging Platforms

Two new messaging platforms join the family:

  • LINE — critical for Japan, Korea, and Taiwan, wired into the LINE Messaging API.
  • SimpleX Chat — privacy-focused decentralized messenger with no user IDs, now a first-class platform.

That brings Hermes to 22 messaging platforms total. Wherever you and your team chat, the agent can be there.

Related PRs: #23197, #26232.

9. /handoff — Live Session Transfer

Switching models or personalities mid-conversation used to mean losing context or starting over. Now /handoff migrates your entire active session — messages, tool calls, context — to a target model, persona, or profile, live, without dropping anything.

# Hand off from a fast model to a deep-reasoning one
/handoff deep-reasoning

# Switch profiles for different parts of a task
/handoff production-profile

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

Related PR: #23395.

10. Native Button UI for clarify on Telegram & Discord

When the agent uses the clarify tool to ask you a multiple-choice question, it now shows real platform-native buttons on Telegram and Discord instead of asking you to type back the option number. Tap the button, the agent gets your answer. Especially nice on mobile where typing option numbers is painful.

Related PRs: #24199, #25485.

11. Discord Channel History Backfill

When Hermes joins a Discord channel or thread for the first time, it now reads the recent message history so it knows what’s been discussed before it responds. No more “what are we talking about?” — the agent walks in with the same context everyone else already has on screen.

Enabled by default. Related PR: #25984.

12. LSP Semantic Diagnostics on Every Write

When the agent uses write_file or patch, Hermes now runs a real language server against the edited file and surfaces any new errors back to the agent before the next turn. Type errors, undefined symbols, missing imports — caught immediately.

This is a significant upgrade from v0.13.0’s syntax-level linting, because it’s actual semantic analysis from a real language server. The agent catches its own mistakes before they compound.

Related PRs: #24168, #25978.

13. Per-Turn File-Mutation Verifier

After every turn that wrote or edited files, the agent now gets a short footer summarizing exactly what changed on disk — file paths, line counts, the actual delta. The agent catches its own mistakes when a write didn’t land or got silently overwritten, instead of confidently telling you “I added the function” when the file wasn’t actually saved.

Related PR: #24498.

14. vision_analyze Returns Raw Pixels

When you point the agent at an image with vision_analyze and the active model can actually see (GPT-5, Claude, Gemini, Grok-vision), Hermes now passes the raw pixels straight to the model instead of converting them to a text description first. You get the model’s actual visual reasoning instead of a degraded text-summary round-trip.

Related PR: #22955.

15. Unified video_generate with Pluggable Backends

One tool, any video model. Hermes ships with the obvious backends already, but you can drop in a new video provider as a plugin without touching core. When a new video model lands next month, it can be a one-file plugin instead of a fork.

Related PR: #25126.

16. computer_use CUA-Driver — Works with Non-Anthropic Models

Computer-use (the agent controlling your mouse and keyboard to drive GUI apps) used to be locked to Anthropic’s SDK. The new cua-driver backend works with non-Anthropic providers too, has proper focus-safe operations, and refreshes itself on hermes update. Now any vision-capable model can drive your desktop.

Related PRs: #21967, #24063.

17. Clickable URLs in Any Terminal

Links in agent output are now real OSC8 hyperlinks with hover-highlight in any terminal that supports them (iTerm2, Kitty, Ghostty, modern Windows Terminal, etc.). Click to open in your browser — no more copy-paste-trim of long URLs from the transcript.

Related PRs: #25071, #24013.

18. Native Windows Support (Early Beta)

Hermes now runs natively on cmd.exe and PowerShell without WSL. A full PowerShell installer handles MinGit auto-install, Microsoft Store Python stub detection, and the foreground Ctrl+C dance. Over 40 Windows-specific fixes landed to make the basic loop work end-to-end on a clean Windows box. Still labeled “early beta” — rough edges remain — but the core path works.

Related PR: #21561.

New Providers & Model Improvements

Feature Description
xAI Grok OAuth SuperGrok subscription integration with OAuth login
NovitaAI New provider for open-source model hosting (Llama, Qwen, DeepSeek, etc.)
OpenRouter Pareto Code Router min_coding_score knob — routes to the cheapest model that meets your quality bar
Codex App-Server Runtime Optional runtime driving OpenAI’s Codex CLI under the hood, with session reuse, automatic retirement of wedged sessions, and OAuth refresh classification
Qwen Cloud Alibaba Cloud renamed to Qwen Cloud (config backwards-compatible)
NVIDIA NIM Billing origin header support

New Optional Skills

Nine new optional skills join the ecosystem:

Skill Description
Hyperliquid Perp and spot trading via SDK and REST API
Yahoo Finance Live market data, fundamentals, historicals
api-testing REST and GraphQL debug recipes
Unified EVM Multi-Chain One skill covering Ethereum + L2s + Base
darwinian-evolver Evolutionary prompt and skill tuning
osint-investigation OSINT recipes for people, domains, and organizations
pinggy-tunnel Expose local services to the public internet
watchers RSS / HTTP JSON / GitHub polling via cron no_agent mode
Notion Full overhaul for the May 2026 Developer Platform

Additionally, huggingface/skills is now a trusted default tap in the Skills Hub — community skills hosted there are installable directly from hermes skills with no extra config.

Related PRs: #23582, #23583, #23590, #25299, #26760, #26729, #26765, #21881, #26612, #26219.

Plugin & Extension Improvements

ctx.llm — plug into the agent’s model pipeline. Plugin authors can now make LLM calls through the active provider and credentials — no manual client wiring needed. Plugins get the same model-routing and auth plumbing the core agent uses.

tool_override — swap out built-in tools. A plugin can cleanly replace a built-in tool with its own implementation. This closes a long-standing community request (#11049).

Shareable profiles via git. Distribute entire Hermes profiles (config, skills, memory) through git repositories — great for team standardization.

Zed ACP Registry. Hermes is now listed in Zed’s Agent Client Protocol registry. Install with one click via uvx, no npm dependency.

Related PRs: #23194, #26759, #20831, #26079, #26120, #26234.

New Built-in Tools & Commands

Tool / Command Description Usage
x_search Search X (Twitter) timeline, threads, and posts Ask the agent to search X directly
hermes proxy OpenAI-compatible local proxy backed by OAuth providers hermes proxy --provider anthropic --port 8080
/handoff Live session transfer to a different model/persona/profile /handoff deep-reasoning
/subgoal Append success criteria to a running /goal /subgoal add unit tests for the new module
/sessions Browse and resume previous sessions in the TUI /sessions
Brave Search Free-tier web search provider Configure in web-search settings
DDGS DuckDuckGo scraper — free, no API key needed Configure in web-search settings

Improvements

Messaging Platforms

  • Telegram: native draft streaming (sendMessageDraft, Bot API 9.5+), notification mode, guest mention mode, split oversized edits instead of silent truncation
  • Discord: thread_require_mention for multi-bot threads
  • Slack: !cmd as alternate prefix for slash commands in threads
  • WhatsApp: surface quoted reply metadata from Baileys
  • Feishu: native update prompt cards
  • Google Chat: repair setup prompt imports, honor relay-declared sender_type
  • Gateway: keeps running when individual platforms fail, per-platform circuit breaker with /platform command

TUI & CLI

  • /sessions slash command for browsing and resuming previous sessions
  • Segment turns with visual rule separators
  • Support attaching to an existing gateway
  • Markdown links resolved to readable page titles
  • Width-aware markdown table rendering with vertical fallback
  • YOLO mode warning in banner and status bar
  • Confirm prompt for destructive slash commands

Kanban (Multi-Agent)

  • specify — auxiliary LLM fleshes out triage tasks
  • Orchestrator board tools: kanban_list + kanban_unblock
  • stranded_in_ready diagnostic for unclaimed tasks
  • Deduplicated notifier delivery via atomic claim + rewind on failure

Google Workspace

  • Drive write operations + Docs/Sheets create/append

i18n

  • All gateway commands + web dashboard localized, 8 new locales added (16 total)

Performance

  • Cold start: ~19s saved through lazy imports, disk-cached model catalogs, parallel doctor checks, and chat -q skipping the welcome banner
  • Browser: 180x faster browser_console via persistent CDP WebSocket
  • Telegram: tuned cadence + adaptive fast-path for short replies
  • Prompt caching: 1-hour cross-session Claude prefix cache on Anthropic, OpenRouter, and Nous Portal
  • TUI: width-aware table rendering, transcript scrolling during prompts

Security & Reliability

Security Hardening

  • Sudo brute-force blocked: sudo -S attempts classified as DANGEROUS; stdin-fed and askpass-stripped sudo invocations caught
  • Three dangerous-command bypasses closed (inspired by Claude Code’s command-detection work)
  • Tool error sanitization: malicious files or remote services can’t inject instructions through error output
  • SSRF fetch paths covered in skills-hub
  • Dashboard auth required for plugin API routes
  • Supply-chain advisory checker scans every install
  • Kanban comment author sanitization — no caller-controlled author override
  • Quick command output redaction and env sanitization
  • Reduced unnecessary shell=True in subprocess calls

Notable Bug Fixes

  • SQLite: fall back to journal_mode=DELETE on NFS/SMB/FUSE (fixes /resume on network mounts)
  • Codex-runtime: retire wedged sessions + post-tool watchdog + OAuth refresh classification
  • Daytona: migrate legacy-sandbox lookup to cursor-based list()
  • MCP: stop retrying initial auth failures, prevent infinite loops
  • ACP: inline file attachment resources
  • Profiles + banner: exclude infrastructure from --clone-all, fix stale update-check repo resolution

Upgrade

hermes update

If upgrading from a version before v0.14.0, back up your ~/.hermes directory first. After the upgrade, run hermes doctor to verify everything is healthy.

For new installations:

pip install hermes-agent
hermes setup

Full changelog on GitHub

← Hermes Agent Changelog