Hermes Bots for iPhone Duo I didn’t want to simply stretch the iPhone app across a bigger screen. I wanted to actually use the fold Open it and Hermes comes alive across both sides: Agent Stage See your Bot’s pet, current activity, elapsed time, live plan, and subagents beside the conversation. Answer its questions and approve actions directly from the Stage. Every Bot’s pet sits in one row, with whoever needs your attention first. Tap to switch. Two chats at once Open a different Bot or group on each half. Keep talking to one agent while checking in with another. Desk Buddy Stand it partly open like a tiny laptop. Your Bot’s pet and replies appear on top. Big Talk, Stop, and Approve buttons sit below. It works with “Hey Herald” too Your Bot on the outside, too The Hermes Pet widget works in StandBy, including on the outer screen. I think the foldable form factor makes sense for AI agents One side to talk to them. One side to watch them work. Herald for iPhone Duo. My unofficial Hermes companion, built around the fold

My Hermes Bots app can now connect to multiple Hermes setups from one phone This was one of the features I wanted most - Home server - VPS - Cloud instance - Different VPNs or networks - Local Add multiple connections, switch between them, and access the Bots running on each one without reconfiguring the app every time Your Hermes agents don’t have to live on one machine or one network anymore One phone. Multiple Hermes setups. All your Bots in your pocket. The unofficial Hermes mobile app keeps getting more useful Android + iOS

Hermes just made its own codebase dramatically cheaper for agents to understand The cleanup is not just for humans It directly reduces how much context agents burn while reading the repo. Some of the numbers are wild: • Median defining file size down 66% • Average lookup tokens down 55% • Multi-window reads down 71% • Tasks hitting files too large for a 128k context: 1,589 → 0 • Total tokens if every task read the whole file: 985M → 361M That is roughly 63% less context for the same work Cleaner code for humans Cheaper, faster reasoning for agents With 3,000+ contributors working on Hermes, this kind of cleanup matters a lot. Teknium keeps optimizing the agent and the codebase the agent has to understand 👏

Hermes Desktop can now see your entire agent network in one place Local agents Remote gateways Hermes Cloud instances Connect them all at once and Desktop gives you one view of every Bot and every chat across the whole setup This is a big deal if your Hermes agents are spread across a laptop, home server, VPS, and cloud Different machines One command center!

This is why I love the Hermes community A user says Group Chat is too slow A few hours later, it’s 4x faster. 35 seconds → 8.6 seconds. Bots now run concurrently instead of waiting on each other, and results push back as soon as they are ready Feedback → ship → improve That loop is what I call fast development

Hermes Agent August Review August started with v0.20.0, The Herald Release, and ended with v0.21.0, The Pantheon Release 2,868 PRs merged in one calendar month I track this project every day, and even I had to double-check that number That is more than 92 merged PRs every single day. Between them came six patch releases, an 844 PR week, and 146 merges on August 31 alone. Here is what defined the month: Hermes found its voice Real-time voice, barge-in, on-device wake words, grounded citations, signed webhooks, and A2A support all landed with Herald. Hermes could speak naturally, verify its claims, connect with other agents, and push events directly into your systems. Agent profiles became a real team Bot Mode turned profiles into named agents with their own role, model, memory, skills, tools, avatar, and persistent chat. Bots can message each other through hermes peer, work across different gateways, join group chats, and hand tasks to specialists. This is where multi-agent stopped feeling like plumbing and started feeling like a team. Cron jobs started remembering Scheduled agents gained persistent memory, continuity between runs, durable notepads, and delivery directly into Bot Chats. Monitor mode can even skip the model completely when nothing has changed, saving tokens and avoiding repetitive alerts. Subagents became manageable Hermes can now list, watch, steer, and stop background agents while they work. You can inspect live progress, redirect a task mid-run, preserve partial results, validate outputs, and track delegation costs. Spawning subagents was already useful. Actually managing them changes everything. MCP and browser automation matured Desktop gained a proper MCP command center with imports, health checks, usage visibility, and install links. Hermes can also control the browser inside Desktop, use Browser Use as its default backend, run Lightpanda locally, and work through a consent-gated copy of your real browser profile. The foundation got stronger August also brought: • Keyless web search for fresh installs • Better instruction and memory protection • Deeper secret redaction • OS keychain encryption • Safer Windows command approvals • More model providers and catalog options • Six infrastructure patches for Docker, hosted, and fresh installs August was the month Hermes stopped feeling like an agent with integrations It started feeling like an operating system for a team of agents Herald gave Hermes a voice, evidence, and connections Pantheon gave its agents identities, memory, communication, and coordination 2,868 merged PRs 2 major releases 6 patch releases The pace is ridiculous The team and community behind it are even more impressive Can’t wait to see what September will bring

Hermes HUD mode now has a “Pretend I’m working” feature 😂 I type: Pretend I'm working HUD instantly throws a terminal on my screen and starts showing fake code like I’m deep in something important Meanwhile, my actual Hermes agents are doing the real work in the background So if someone walks past my desk... It looks like I’m grinding. Technically, someone is I know this PR will never make it so have fun👇

Hermes Agent v0.21.0 just dropped with nearly 2,500 merged PRs This is The Pantheon Release Since v0.20.0: • 5,800 commits • 2,475 merged PRs • 2,100 issues closed • 760+ contributors The headline is Bot Mode, but this release is much bigger. Here’s what matters most ↓ 1. Bot Mode is now built into Desktop Every Hermes profile can become a named Bot with its own: • Role • Model • Memory • Skills • Tools • Avatar • Permanent chat You can create group chats, mention Bots directly, and watch a team of specialized agents work together. 2. Bots can message each other With hermes peer, one agent can send work to another across profiles and even different gateways. A research Bot can hand findings to a coding Bot, then return the answer to the original conversation. The handoff is durable and visible, not hidden somewhere in a log. 3. Cron jobs finally remember Scheduled agents now have persistent memory, continuity between runs, and their own durable notepad. Your morning briefing can remember what it told you yesterday. Monitoring jobs can also skip the model entirely when nothing changed, saving tokens and avoiding repeated alerts. 4. Subagents are no longer fire and forget Hermes can now: • List running subagents • Read their progress • Steer them while they work • Stop one early • Keep its partial result • Validate its final output • Show the cost of each delegation This is real multi-agent control. 5. MCP management became a command center The new Desktop experience brings MCP servers and the catalog into one place. You can import servers, monitor health, catch authentication problems early, inspect usage, and see estimated schema token costs. No more digging through config files to manage a fleet of MCPs. 6. Hermes can control its own Desktop browser The in-app browser is no longer something Hermes can only observe. It can now navigate, click, and read pages directly inside Desktop. Ask it to explore documentation, test a web app, or investigate a page and watch it happen. 7. Huge CLI, model, security, and reliability upgrades • Fuzzy command and model pickers • Richer /status with live performance metrics • Global emergency stop • Six new model providers • More models across the catalog • Protected instructions, skills, and memory files • Stronger secret redaction • Better Windows command approvals • More reliable sessions, updates, messaging, and recovery This is not a normal feature release Hermes is turning individual agents into a persistent team that can remember, communicate, delegate, monitor, and operate together. Massive work by NousResearch and the entire Hermes community. 🪽

Hermes merged 60 PRs on August 30 The ones worth knowing: • Credential vault lives in Settings, with agent deep link prefill • Real profile browsing: Brave Origin detection, macOS signed in, headless (no focus steal), toggle under Capabilities to Tools to Browser • Desktop MCP OAuth sign in works against remote backends • /loop runs its first iteration immediately • /plan is a built in command that survives platform menu caps • Telegram: every command and skill searchable via @botname inline picker • CLI and TUI status show cache hit %, latency, and t/s with field toggles • System prompt always rebuilds at compaction commit, so updates reach long lived sessions • Skill activation drops repeated absolute paths (down 31% load on reference heavy skills) • Copied session cookies are no longer world readable • Unattended webhook/API sessions no longer stall 300s on dangerous command approval • Lean compaction is one auxiliary request again (7 to 11 min digest stalls gone) Here is the full list: DESKTOP IMPROVEMENTS (13 PRs) • feat(desktop): manage the credential vault from Settings, with agent deep link prefill • fix(computer use): stop launching retired browser grant runtimes • fix(tui): render nested todo subtasks via the parent field • fix(desktop): remote bot replies no longer die at the 30s relay deadline (#93911#, salvage #93965#) • feat(browser): Brave Origin no longer misdetected, real profile browsing + CDP connect (salvage #41505#) • fix(todo): Tasks panel stays live across merges, reconnects, and hidden tool progress (salvage #97815#, #97811#) • fix(browser): real profile browsing on macOS no longer opens signed out or hangs (salvage #96763#) • feat(mcp oauth): Desktop MCP OAuth sign in now works against remote backends • fix(cron): Desktop's advertised natural schedules now parse (#51975#, salvage #51598# #52504#) • feat(desktop): the real profile browsing toggle users couldn't find, Capabilities to Tools to Browser • feat(tui): status rule shows cache hit %, latency, and t/s with shared field toggles (extends #98250#) • fix(browser): real profile browsing runs headless, no focus stealing window • fix(todo): unversioned tool.start merges again after desktop resume (#98304#, salvage #98306#) DEVELOPER TOOLS & PLUGINS (22 PRs) • fix(providers): register Alibaba China + Token Plan providers, 'Unknown provider' despite catalog entry (#73265#) • fix(memory): keep Mem0 OSS OpenAI requests direct • feat(loop): /loop runs its first iteration immediately (salvage #97958#) • fix(skills): skill_view directory file_path + skill_manage categorized name resolution • fix(prompt): sync DEFAULT_SOUL_MD with the #95681# identity rewrite • docs: sync stale /background references with the /bg + /btw split • fix(tools): restore setup_mcp's never hand edit instruction • fix(delegation): subagents on an explicit base_url keep their provider's request_overrides (#65035#, salvage #65054#) • feat(skills hub): impeccable joins the optional skills catalog, content pulled live from upstream • feat(cli): status bar gains cache hit %, latency, t/s, and per field toggles (salvage #92696#, #39760#, #97970#) • hermes model shows curated lists for all six Alibaba tiers (salvage #77848#) • Background review lands skill patches again, read tools whitelisted, cache parity intact (#61521#, #39996#) • docs(computer use): browser route and grant references match the shipped tool (salvage #96156#) • Qwen Cloud picker gains qwen3.8 max + 4 more missing models (salvage #87808#) • feat(delegation): delegation.request_overrides config key sends per child request settings on every resolution branch (salvage #90953#) • feat: /plan survives the platform command menu caps as a built in command (closes #67264#, #36821#, salvage #67292#) • Background review accepts profile configured extra tools via opt in whitelist (closes #44672#, salvage #82146#) • fix(skills hub): skill installs survive dangling refs, sibling links, glob prose, and unpinned fetches (salvages #66828# #82236# #90105# #96336#) • feat(compaction): system prompt always rebuilds at the commit boundary, updates finally reach long lived sessions • fix: skill activation no longer repeats the absolute path on every supporting file line (down 31% load cost on reference heavy skills) • refactor(skills): move merge reconciler to optional catalog as agent merge conflict arbiter • refactor(skills): shipped set slim, 15 to optional, github 6 way merge, pdf absorbs OCR, channel gated teams pipeline (index down 26%) SECURITY & SECRETS (4 PRs) • Anthropic OAuth: shared credential refreshes stop burning single use tokens; dashboard PKCE removed (salvage #87891#) • fix(browser): copied session cookies are no longer world readable (#96729#) • fix(security): env vars with mid name API/KEY/TOKEN no longer flag as exfiltration (salvage #63994#) • fix(approval): unattended webhook/API sessions no longer stall 300s on dangerous command approval (#37284#, salvage #37317#) INTEGRATIONS & CHAT (5 PRs) • fix(telegram): recover exhausted request pool • fix(runtime): provider switch, fallback, and recovery keep request settings in sync (#75091#, salvage #75139# #60062#) • fix(gateway): custom provider extra_body survives gateway turns and /model switches (#54922#, salvage #39429# #52432# #53765#) • fix(telegram): pinned skill commands survive the menu cap (salvage #82516#) • feat(telegram): every command and skill is searchable via @botname, inline picker bypasses the menu cap STABILITY & FIXES (16 PRs) • fix(cli): slow /handoff transfers no longer misreported as "gateway not running" • 3x fmt(js): npm run fix auto fix • fix(cron): 'every hour' and 'every day' schedules no longer error (salvage #97783#) • fix(cron): scheduled jobs no longer lose custom provider request settings (salvage #56876#) • fix(cron): bare '30m' schedules recur as documented; 'in 30m' is the one shot (salvage #78928#) • fix(cron): repeat='forever' no longer breaks job updates (completes the string repeat class) • fix(compression): stale todo snapshots no longer accumulate or resurrect finished plans (salvage #97604#, #89910#) • fix(compression): native Codex compaction no longer fires at the stale 200K default on 900K windows (#88695#, salvage #88722#) • fix(compression): compacted native Codex sessions no longer false trigger 600s local compression mid turn (#96995#, #97602#, salvage #97006#) • fix(native compaction): image only user messages survive checkpoint pruning (salvage #91557#, folds #98345# test) • fix(codex): compaction only turns no longer die after 3 continuation attempts (salvage #98008#) • fix(compaction): native capability survives same provider /model switches and gateway resume (salvage #94036# + #97292#) • fix(compression): native compaction settings hot apply to open sessions (salvage #96740#) • fix(compression): lean compaction is one auxiliary request again, 7 to 11 min digest stalls eliminated (#96603#)

I want to hear from you, Hermes builders I spend a lot of time building with Hermes, testing new features, sharing workflows, and helping spread the word But I don’t want to keep guessing what the community wants What would you like to see more of? Hidden gems? Tutorials? Real workflows? Live builds? Beginner guides? Open source projects? Community challenges? Tell me what would actually help you I’m listening 👇

Hermes merged 57 PRs on August 29 The ones worth knowing: • /btw now keeps conversation context. /background is retired in favor of /bg • HUD clicks, links, and right-click work on Linux X11 • Task lists nest • grill-me: adversarial plan interview • Desktop: download any file from a preview card • Kanban boards export, import, rename, and delete • New providers: Nebius Token Factory, Ramp Router, Tencent TokenPlan + hy4-preview • /busy is now on gateway platforms • Slack streamed finals unfurl correctly • Voice-bubble transcode across platforms • Ghostty-level pets in the CLI • Optional skills: decision-questionnaire and setup-wizard-generator Here is the full list: DESKTOP IMPROVEMENTS (8 PRs) • feat(desktop): Download button on preview file cards • fix(desktop): ::preview inline widgets work on remote/URL connections • fix(desktop): board switcher crashed on every render • fix(desktop): HUD clicks, links, and right-click on Linux X11 • fix(desktop): MEDIA: any-file delivery — preview card + Download button for non-media files; guidance aligned • fix(desktop): reserve space for pane tab close button • Kanban board export / import, plus board rename and delete in the desktop switcher • refactor(desktop-tools): consolidate preview + project, diet the desktop_ui suite (3,861 → 2,293 tok/call, −41%) VOICE & REAL-TIME (1 PR) • fix(prompt)+feat(gateway): platform-hint truth pass + universal voice-bubble transcode (all 22 hints source-verified) DEVELOPER TOOLS & PLUGINS (16 PRs) • Add `tencent/hy4-preview` to model pickers • feat(providers): add Nebius Token Factory provider (salvage #28253#) • feat(providers): add Ramp Router ( provider plugin (salvage #93548#) • feat(providers): add Tencent TokenPlan + hy4-preview default (salvage #96939#) • feat(cli): Ghostty-level pets in the interactive CLI • feat(optional-skills): decision-questionnaire — async questionnaires for blocked decisions • feat(optional-skills): setup-wizard-generator — bash wizards for human-only setup steps • feat(skills): add optional grill-me skill — adversarial plan interview (salvages #43694#) • feat(skills): grill-me becomes plan-interrogation with frontier-rounds interviewing • feat(todo): task lists now nest — subtasks via a ~45-token parent field • /btw answers now ride the background-review warm-cache fork (full context, cache-read cost) • /btw answers side questions with conversation context; /background retired in favor of /bg • feat(prompt): DEFAULT_AGENT_IDENTITY — behavior spec replaces trait list (concise-by-default) • feat(system_prompt): two-line conversation clock — anchored start + rebuild-day line (salvages #96224#) • docs(user-stories): add 65 new community stories, newest first • chore: map provider-salvage contributor emails (Neel49, amrrs) SECURITY & SECRETS (4 PRs) • fix(mcp-oauth): avoid locking across resource requests • fix(teams): authenticate Bot Framework connector attachment downloads (pasted images 401) • fix(vertex): pick up rotated service-account files — signature-keyed creds cache • refactor(vertex): fold _creds_cache_key into _sa_snapshot (post-#97701# simplify)

Hermes merged 81 PRs on August 28 DESKTOP IMPROVEMENTS (15 PRs) • fix(desktop): stop the workspace pane and pooled sockets flickering at idle • fix(desktop): let the HUD drag onto another monitor • fix(desktop): 'This device' Capabilities scope reaches the local machine again under a remote regist... • fix(paths): display_hermes_home renders POSIX separators on Windows (kills ~/AppData\Local\hermes ch... VOICE & REAL-TIME (3 PRs) • fix(desktop): stop session-switch transcript flicker (supersedes #96130#, #82899#) • fix(update): gate the fork-upstream prompt for --yes and non-tty runs • fix(desktop): Bot Chat reopen refetches the transcript instead of the idle snapshot (#96183#, salvage... DEVELOPER TOOLS & PLUGINS (17 PRs) • feat(a2a): client tools config-gated — disabled unless enabled (−561 tok/call on unconfigured instal... • fix: strict positional providers no longer 400 on histories with displaced or replayed tool results ... • fix(agent): tool-pair sanitizers — whitespace ids pair, stale empty tool_calls drop, orphan logic un... • fix(desktop): a malformed mcp_servers entry no longer blanks the whole workspace (salvage #94338#) SECURITY & SECRETS (2 PRs) • fix(agent): 413 recovery measures bytes, not token estimates (salvage #88960#) • fix(browser): CDP screenshots survive redaction byte-identical — no more provider 400 invalid_image ... INTEGRATIONS & CHAT (5 PRs) • fix(gateway): /handoff works on multi-profile installs (salvage #91217#, fixes #91216#) • fix(gateway): event loop no longer freezes on hot blocking sites; ASYNC lint gate stops the class (s... • fix(gateway): stale model/provider keys no longer survive session model_config writes (salvage #9674#... • fix(cron): manual-run prompt no longer dropped on the relay-fronted gateway forward STABILITY & FIXES (39 PRs) • fmt(js): npm run fix auto-fix • refactor(patch): V4A gated to OpenAI-family mains — replace-only base (−149 tok/call for every other... • fix(install): locked uv sync sees project config in BOTH Unix installers without re-exposing ambient... • fix(install): fresh installs keep hash-verified deps and Node 24.0-24.10 no longer dies at npm (salv... 81 new features and fixes

Hermes3D now has a 2D pixel office mode Same agent command center Completely different vibe Now you can watch your Hermes agents work inside a full pixel office with: • Team areas • Meeting rooms • Phone booths • Ops corner • Server room • Lounge • Gym • Game room The goal is still the same Make your AI team feel visible, understandable, and actually fun to manage 3D when you want immersion 2D pixel mode when you want the classic office sim feel Hermes3D keeps getting more fun

Hermes merged 88 PRs on August 27 DESKTOP IMPROVEMENTS (16 PRs) • fix(desktop): Cronjobs pane collapses to a ribbon instead of vanishing (salvage #95956#) • fix(desktop): spare durable rows when a sibling backend still holds the lease (supersedes #65059#) • fix(model): coerce numeric custom provider names so Desktop can assign them • fix(serve): write machine sentinels to real fd 1 so the desktop handshake survives the stdout redire... VOICE & REAL-TIME (6 PRs) • fix(desktop): gate transcript budget cap so Show earlier works (supersedes #87686#) • feat(wecom): native reply streaming — typing bubble, per-turn isolation, dedup-safe delivery (salvag... • fix(sessions): don't let the empty-session sweep delete an archived transcript • fix(desktop/sessions): legacy NULL-owner session migration — single-match owner backfill + read-only... DEVELOPER TOOLS & PLUGINS (18 PRs) • fix(cli): keep journey labels readable • fix(desktop): a Bots-pane click keeps the open bot chat focused (#96062#, salvage #96120#) • state.db journal mode has one owner: repair re-applies it, delegation guests stop flipping it (salva... • fix(gemini): embed images in Gemini 3.x for multimodal tool results SECURITY & SECRETS (2 PRs) • chore: AUTHOR_MAP — shauneccles (#95433#) + fedosis (#94996#) • fix(desktop): recover cloud auth through portal INTEGRATIONS & CHAT (11 PRs) • fix(gateway): size systemd TimeoutStopSec from the full stop budget incl. the cron drain floor (salv... • fix(tui-gateway): a reconnected session no longer dies to the old socket's disconnect teardown (#771#... • fix(gateway): keep sessions.json route when startup recovery succeeds with the same session id (salv... • fix(gateway): preserve exception type when error string is empty — stop duplicate delivery on httpx ... STABILITY & FIXES (35 PRs) • fix(cron): make a recurring job stuck in state=error recoverable again • fix(compression): dedupe current-turn rows when rotation splits the session mid-turn • fix(compression): retry a stalled summary on the fallback chain (#78981#) • fix(bot-mode): deliveries no longer break on Docker/remote terminal backends (salvage #95603#)

Hermes HUD mode just became my live stock analyst I opened a chart and asked Hermes to analyze what it was seeing Instead of only explaining it in text, HUD drew directly over my screen: • Resistance • Support • Trend direction • Current price context So now Hermes can look at the same chart I’m looking at and annotate the analysis in real time. This is exactly what I wanted HUD mode to become. Not another chat window. An AI layer on top of whatever I’m already doing 👀

Hermes can now browse the web already logged in as you Turn on Real Profile Browsing and Hermes creates a managed copy of your active Chrome, Edge, Brave, or Chromium profile. Your cookies Your logins Your preferences Hermes never opens your live browser profile directly. It uses a separate snapshot, refreshes the login data for new sessions, and deletes the copied profile when you turn the feature off. That means your agent can finally work inside the tools you already use without asking you to log in every time. This is powerful, so it is off by default. Enable it only when you actually want Hermes acting as you. Settings → Browser → Use My Real Browser Profile Browser automation just became personal