Release History
Every release, every improvement. Full version history sourced from the Hermes Agent GitHub.
The Quicksilver Release makes Hermes substantially faster and more durable across CLI, TUI, Desktop, gateways, and background agent workflows.
- About 80% lower first-turn time-to-first-token, plus faster Desktop and TUI streaming
- Subscription and top-up management in the terminal, with matching Desktop billing settings
- Smart approvals by default, user-defined deny rules, and reason-aware /deny
- Bitwarden and 1Password integrations through pluggable secret sources
- Live, durable subagent transcripts and restart-safe final-message delivery
Same-day v0.18.1 follow-up that makes tagged-release Docker builds reliable by using the published WhatsApp Baileys dependency.
- WhatsApp Baileys uses published 7.0.0-rc13
- Reliable dependency installation
- Tagged-release Docker build fix
Infrastructure-focused stability tag that packages the post-v0.18.0 fix window for Docker images, hosted deployments, and PyPI installs.
- Installer and updater self-healing
- Dashboard and gateway fixes
- WhatsApp pairing, MCP, and provider stability
The Judgment Release clears the tracked P0/P1 backlog and adds evidence-led completion, first-class Mixture-of-Agents, visible self-improvement, background fan-out, Desktop coding Projects, and production-focused gateway controls.
- All tracked P0/P1 issues and PRs closed at release
- First-class Mixture-of-Agents model presets
- Verification evidence and completion contracts for /goal
- Learn, journey, and Desktop memory graph
- Projects, background fan-out, and scale-to-zero gateway support
The Reach Release expands Hermes across messaging, desktop, browser administration, automation, image editing, skills, and memory workflows.
- iMessage via Photon and Raft agent network
- Background subagents and image editing
- Automation Blueprints and browser profile builder
- Desktop productivity, themes, and remote-gateway improvements
- Skills Hub rehaul and atomic memory batch operations
The Surface Release: a native desktop app, browser admin panel, remote gateway connections, Simplified Chinese desktop UI, leaner default skills, trusted NVIDIA skills, fuzzy model picker, and /undo.
- Native desktop app for macOS, Linux, and Windows
- Full browser-based administration panel
- Remote gateway sign-in with OAuth or username/password
- Simplified Chinese desktop UI
- Fuzzy model picker and /undo across interfaces
v0.15.2 bugfix release that ships bundled plugin.yaml manifests in the wheel and source distribution.
- Bundled plugin.yaml manifests included in Python packages
- Wheel and sdist packaging fix
- Follow-up release after v0.15.1
The Patch Release: fixes the dashboard reload loop in loopback mode, Docker insecure opt-in handling, MCP command resolution, skills catalog loading, and kanban worker shutdown.
- Dashboard 401 reload loop fixed
- Docker dashboard --insecure now requires explicit opt-in
- MCP bare commands resolve correctly in Docker
- Full skills.sh catalog restored
- Kanban worker SIGTERM handling fixed
The Velocity Release: major run_agent.py refactor, expanded multi-agent Kanban, faster cold starts, free and instant session_search, promptware defenses, Bitwarden Secrets Manager, and new image generation providers.
- run_agent.py reduced from 16k lines into focused agent modules
- Kanban expanded into a multi-agent platform
- Cold-start and per-turn performance wave
- session_search rebuilt to be no-LLM and much faster
- Promptware defense, Bitwarden, Krea, FAL, and ntfy additions
The Foundation Release: PyPI install, native Windows beta, xAI Grok via SuperGrok OAuth, OpenAI-compatible local proxy, x_search, Microsoft Teams, lazy installs, and broad platform expansion.
- pip install hermes-agent support
- OpenAI-compatible local proxy for OAuth providers
- xAI Grok, x_search, and Microsoft Teams integration
- Lazy dependency installs and supply-chain hardening
- LINE, SimpleX Chat, Windows beta, and new optional skills
The Tenacity Release: durable multi-agent Kanban, persistent /goal, Checkpoints v2, gateway auto-resume, no_agent cron mode, stronger security defaults, Google Chat, provider plugins, and seven i18n locales.
- Durable multi-agent Kanban with retries and recovery
- Persistent /goal and Checkpoints v2
- Gateway auto-resume and no_agent cron watchdog mode
- Security wave with redaction on by default
- Google Chat, provider plugins, and seven i18n locales
Autonomous background Curator, substantially upgraded self-improvement loop, four new inference providers, Microsoft Teams integration, and 57% faster TUI startup.
- Autonomous background Curator for continuous skill improvement
- Substantially upgraded self-improvement loop
- 4 new inference providers added
- Native Microsoft Teams integration
- 57% reduction in TUI cold start time
New Ink-based TUI, pluggable transport architecture, native AWS Bedrock support, and GPT-5.5 via Codex OAuth.
- New Ink-based terminal UI with improved rendering
- Pluggable transport architecture for extensibility
- Native AWS Bedrock provider support
- GPT-5.5 access via Codex OAuth
Tool Gateway enabling web search, image generation, text-to-speech, and browser automation for Nous Portal subscribers.
- Tool Gateway for Nous Portal subscribers
- Web search capability
- AI image generation
- Text-to-speech output
- Browser automation support
Mobile support via Termux/Android, iMessage and WeChat additions, and Fast Mode for OpenAI and Anthropic providers.
- Native Android support via Termux
- iMessage gateway added
- WeChat messaging integration
- Fast Mode for OpenAI and Anthropic
Native Google AI Studio support, live model switching, self-optimized guidance, and centralized logging.
- Native Google AI Studio provider
- Live model switching without restart
- Self-optimized guidance system
- Centralized structured logging
Pluggable memory providers, credential pools, Camofox browser integration, and secret exfiltration blocking.
- Pluggable memory provider architecture
- Credential pool management
- Camofox browser integration
- Secret exfiltration protection
User profiles, MCP server mode, Docker container support, and Telegram webhook mode.
- User profile system
- MCP server mode for external tool access
- Docker container runtime support
- Telegram webhook mode
Nous Portal with 400+ models, Hugging Face provider, Telegram Private Chat Topics, and Nix flake support.
- Nous Portal integration with 400+ models
- Hugging Face model provider
- Telegram Private Chat Topics
- Nix flake for reproducible installs
300 merged PRs in one week — streaming output, native browser tools, and plugin system.
- 300+ community PRs merged in a single week
- Streaming output support
- Native browser tool integration
- Plugin system architecture
Unified streaming, plugin architecture, native Anthropic provider, and voice mode.
- Unified streaming across all providers
- Foundational plugin architecture
- Native Anthropic provider (no API wrapper)
- Voice mode for audio interaction