Hermes Agent Release Notes

Patch release: ~323 PRs and ~746 commits since v0.20.4. Bot Mode group chats finally accept files (PDF/image drag & drop), old conversations fold into one-line summaries, and blob avatars arrive; hermes worktree list/prune safely reclaims accumulated worktrees and merged branches; hermes update --plan plus structured update receipts make upgrades auditable; cron jobs gain persistent memory and per-job reasoning effort; and opencode-free becomes a truly keyless free model in the /model picker.

Patch release: ~74 PRs and ~146 commits since v0.20.3. A macOS Glass translucency mode you can actually read through, a permanently tabbed SESSIONS|BOTS sidebar with per-bot hide/unhide, advisory NVIDIA SkillEvaluator scans on skill installs, cron media-send hardening, SessionDB contention fixes, an honest hermes update, and native Kanban notifications.

Patch release: ~125 PRs and ~250 commits since v0.20.2. The cron scheduler learns to heal itself (EMFILE, stale claims, wedged jobs, misfire catch-up), MCP speaks the 2026-07-28 stateless protocol, CommandCode joins as a first-class provider, and /worktree + /rollback make agent edits reversible.

Patch release: ~397 PRs and 967 commits since v0.20.1. Desktop gains a multi-connection registry (local, remote gateways, Hermes Cloud, SSH), MCP gets hermes:// one-click installs and background health checks, terminal breadcrumbs make -c per-terminal, and image upscaling becomes opt-in.

Patch release: ~656 PRs and 1,444 commits since v0.20.0. Messaging platforms grow to 18, Windows setup stops hanging, Anthropic prompt caching returns, state.db fully self-heals, native live cards on Slack.

Real-time conversational voice with barge-in. On-device wake words, hands-free control. Voice on every platform — WhatsApp, Feishu, DingTalk, LINE, QQ, WeChat. Trusted research with grounded citations and fact-checking. Agent-to-Agent protocol A2A v1.0. Signed outbound webhooks. Desktop platform wave — Artifacts, plugin SDK, quick-entry, multi-window. CLI power commands: !shell, /init, /diff, /context, /focus. Tool self-recovery. Smarter context compression. Cold start ~1.8s. 650+ contributors.

Cold-start latency cut by 80% to 0.9s. Reasoning streaming live by default. Smart approvals with LLM-based command review. Bitwarden & 1Password secret sources. Live subagent transcripts with durable background delegation. Delivery-obligation ledger for crash-proof responses. Profile-based message routing. GPT-5.6, grok-4.5, kimi-k3, claude-sonnet-5 support. Desktop 20+ PR speed overhaul. Terminal billing via /subscription and /topup.

100% P0/P1 resolved (496 issues + 196 PRs). Mixture-of-Agents as a first-class pickable model with live-streamed reasoning per reference model. Agent self-verification with /goal completion contracts and pre_verify hooks. /learn skill distillation, /journey memory timeline, background fan-out subagents, desktop coding Projects, scale-to-zero gateway with drain coordination, Google Vertex AI native OAuth2 support, and 10 messaging platform updates.

iMessage via Photon Spectrum with no Mac relay. Background/async subagents. Image-to-image editing. Automation Blueprints without cron syntax. Desktop app with rebindable shortcuts, subagent watch windows, VS Code themes. Official WhatsApp Business Cloud API. Cursor Composer model via xAI Grok. Full profile builder + Skills Hub rehaul. Memory batch atomic operations. God-file refactor: cli.py 3297→954 lines, gateway/run.py 19157→15870 lines.

A brand-new native desktop app for macOS/Linux/Windows with drag-and-drop, in-app updates, and multi-profile remote gateway connections. Full web dashboard administration panel, Simplified Chinese translation, NVIDIA/skills trusted tap, leaner default skill set, fuzzy model picker everywhere, /undo command, Quick Setup via Nous Portal, and 399 issues closed (2 P0, 62 P1).