Hermes-Workspace v2.3.0 — HermesWorld integration, Agent View, Dashboard polish

Hermes-Workspace v2.3.0 — HermesWorld integration, Agent View, Dashboard polish

A focused follow-up to v2.2.0 with three new features in the workspace shell and a wave of bug fixes that unblock fresh installs.

46 commits since v2.2.0 · 22 PRs merged in the final push · 4 issues closed · queue down 50%.

---

✨ New features

🌐 HermesWorld integration

The playable agent MMO — hermes-world.ai — now embeds directly inside Workspace via iframe, with the workspace nav preserved around it. Sign in with Google or email, reserve your founder username, and you're in. Real users started signing up the moment the landing went live (13 founder usernames claimed in the first hour).

• /play/ embed with workspace chrome around the playable runtime
• Wave Chat RPG panels — Inventory, Quest, Map, Founders Vault as React components
• Speech bubbles + toasts in the in-world chat
• Photosensitive safety mode — flash/strobe reduction toggle
• Mobile bundle splitting for faster load on phones
• HermesWorld brand asset pack — logo lockups, social previews, in-game art
• MJ-generated NPC + zone art wired into the playground
• Public name reservation flow at /reserve with founder ranking

VITE_HERMESWORLD_ENABLED=false lets you opt out of the link in the sidebar.

👤 Agent View panel

A new side panel in chat that shows the active agent, queue, history, and a usage meter wired to real provider quota.

• Avatar reflects real activity state — idle / thinking / speaking, sourced from the chat-activity store, no fake demo data
• Two-tier avatar picker — Hermes default + 9 Greek god PNGs (Athena, Apollo, Artemis, Iris, Nike, Pan, Chronos, Eros, Hermes), plus an emoji avatar tier
• Real provider quota tracking in the usage meter
• Smooth slide+fade transitions on the desktop panel
• Mobile sheet variant for small screens
• Enterprise chrome — borderless inner sections, no demo fallback when the gateway returns zero agents

📊 Dashboard

Workspace dashboard now mirrors the native Hermes Agent dashboard at :9120, aggregated server-side in a single round trip that proxies six native endpoints with per-section graceful fallbacks.

New cards:

• System Status Strip — gateway pill + active-agents count + warning chip on restart_requested
• Platforms card — connected platforms with per-platform state pills
• Cron Summary — scheduled / paused / running counts + next-run countdown, deep-links to /jobs
• Achievements ribbon — three most recent unlocks with tier badges + modal for the full ribbon
• Analytics Summary — top-3 models by tokens with proportional bars + real cost (replaces the hardcoded $5/M estimate)
• Provider Mix donut + cache efficiency tile
• Velocity + Cost Ledger menu items
• Contextual Operator Tip card

Vanilla installs still render a usable dashboard — every section resolves to null independently if its provider isn't available.

---

🛠 Patches & fixes (22 PRs merged)

Install / runtime

• #360 — Docker compose finally works on fresh installs. hermes-agent was missing command: [gateway, run] so the container started the interactive CLI and exited under up -d. Fixed in #385.
• #356 — "Hermes updated" modal stops popping on every refresh. storeNotes() was clearing the seen marker on every poll. Fixed in #386 — only clears when the notes ID actually changes.
• kimi-k2.6 256k context window support (#357)
• Legacy claude-workspace / claude-agent remote aliases so old upgrades don't break the update checker (#359)

Tasks / jobs / cron

• Tasks board auto-detects local vs gateway backend (#361). Probes /api/hermes-tasks and /api/claude-tasks in parallel and picks the one with data. Fixes silent empty boards on installs that store tasks in ~/.hermes/tasks.json.
• Jobs API normalization — always returns {jobs:[]} even when the gateway returns a bare array (#358)

Chat / sessions / agent

• Chat session accounting + local title rename fixes (#350) — phantom agents in sidebar, context meter inflation, local session rename display mismatches all addressed.
• Inline MEDIA:<path> rendering (#349, #330) — local file paths from the agent now show as images in chat, served through a path-allowlisted /api/media route.
• Inline artifact cards in chat (#346, #295)
• Workspace session identity preserved during streams (#310)
• Right-click works on update cards (#286)

Conductor / Swarm / orchestration

• Conductor mission sanitization (#335) — strips Cloudflare 5xx HTML error pages and self-referencing workspace URLs out of mission goals before they hit the model. Defense-in-depth against gateway/proxy failures contaminating agent prompts.
• Conductor falls back when Dashboard mission API is unavailable (#317)
• Swarm tmux session survival check (#341) — when the agent process dies inside tmux (bad PATH, missing API key, etc.), Workspace now surfaces the actual exit output instead of a cryptic "can't find pane" error.
• Hardened workspace swarm prompt submission (#307)
• Operations picker merges Hermes models (#342)
• Chat sends scoped to active workspace (#340)

Files / browser / settings

• Server-side file browser mode clarified (#345)
• HTML preview mode in file browser (#296)
• Groq STT controls in workspace settings (#347)

Themes / UI

• SciFi theme (#320) — full dark/light Tailwind v4 token palette
• Switch toggles show ON/OFF labels with emerald accent (#284)

Auth / gateway

• Codex OAuth tokens bridge to portable-mode chat bearer auth (#332)
• Gateway capability fields reported correctly (#318)
• Production server start path fixed (#308)
• Faster recovery from disconnected state (#275)

Misc

• VT-Capital Guardian OMS cockpit plugin-scoped surface (#364)
• Agora HUD polish to match mockup (#376)
• MseeP.ai security badge added to README (#378)
• Asset generation v2 prompts + ChatGPT prompt batch docs (#370, #375)
• HermesWorld lore bible + player guides (#368)

---

🔄 Issues closed

• #360 Docker compose on fresh install — fixed in #385
• #356 Update modal popping every refresh — fixed in #386
• #319 Workspace + Codex provider compatibility — clarified, gateway is the missing piece
• #294 File browser file system access — already resolved, server-side mode confirmed

📊 Queue movement

• PRs: 24 open → 12 open
• Issues: 10 open → 6 open
• Today's merges: 22 PRs

---

🚀 Upgrade

cd ~/hermes-workspace
git pull
pnpm install
pnpm dev

Or, fresh install:

curl -fsSL https://hermes-workspace.com/install.sh | bash

---

🙏 Contributors this release

@Interstellar-code, @WanderWang, @kadeross, @John-tip, @shoveller, @Fungraphic, @valted81, @kung4lex, @aliasfoxkde, @moto-taka, @kaankirsan, @nazeshinjite, @nixbin-engineering, @CipherFrame, @mseep-ai, @SebuH, @kliebinz-ds, @yasuhidekoizumi-afk

---

Free. MIT. Open source forever.

→ github.com/outsourc-e/hermes-workspace
→ hermes-world.ai
→ hermes-workspace.com

Hermes-Workspace v2.2.0 — HermesWorld

Hermes-Workspace v2.2.0 — HermesWorld

The biggest gameplay-shaped release we've shipped. HermesWorld turns the playground into a full multiplayer 3D environment where every NPC teaches a real Hermes Agent skill. 222 commits since v2.1.3.

✨ HermesWorld

A six-world action-RPG that lives at /playground (now featured as HermesWorld in the workspace sidebar with a gold NEW badge):

• Six worlds — Training Grounds, Forge, Agora, Grove, Oracle, Arena. Each one teaches a real concept (prompts, skills, multi-agent coordination, memory, routing, evals).
• Every NPC = a real skill — Athena, Pan, Hermes, the Knight, and the rest. Talking to them surfaces actual Hermes Agent capabilities.
• Multiplayer out of the box — Cloudflare Hibernation API powers the public WS hub. Any two devices that open the workspace see each other in the same world. HTTP polling fallback keeps it bulletproof on flaky networks.
• Premium presentation — cinematic camera, ASCII NPC portraits, lore loading screen, knight armor models, in-world speech bubbles, focus mode, and a hero canvas agent network on the title screen.
• Quests — Tutorial loop in Training, Forge unlock gate, Diplomacy & Summoning quests, journal upgrades, low-HP cinematics, mobile-friendly layout.

The full demo script and recording kit live at memory/goals/2026-05-03-playground-training-grounds/.

🎨 Themes

• Matrix dark / light (#279, thanks @Interstellar-code) — phosphor green on cinematic black, plus a paper-white companion.

⚙️ Settings & onboarding

• Custom OpenAI-compatible provider UI (#287, @Interstellar-code) — always-clickable Custom card with editable Base URL + API key. Stored in .env as CUSTOM_API_KEY, with base URL in config.
• Provider card verified status (#282, @Interstellar-code) — single-dot precedence and a verified-only green dot. Local providers no longer pretend to be online when they're not.
• Auth Secure-cookie LAN warning (#281, @Interstellar-code) — startup warning when running in production mode behind a non-loopback bind address without HTTPS, since browsers silently drop Secure cookies on plain HTTP.

🌍 i18n

• Japanese (ja) locale (#290, @yasuhidekoizumi-afk) — full translation pass replacing the previous ja: EN fallback, plus regression tests mirroring the existing zh/ru coverage.

🔍 Search

• Recent Searches now real (#292) — replaced the hardcoded placeholder strings with localStorage-backed history (cap 6, dedupe by case-insensitive match). The section hides itself when there's no history yet.
• Chats scope actually finds chats (#291) — the search modal now matches against the API's derivedTitle and preview instead of just the raw friendlyId. Queries like github or workflow will now hit the chat that was about that topic instead of returning "No results found."

🛠️ Dev-server fixes

• Vite connection-status returns the full capabilities map (#285) — the dev server's inline shortcut for /api/connection-status returned a slim {ok, mode, backend} body that silently broke useFeatureCapability and useFeatureAvailable in dev. The shortcut is removed; the real route file (which already caches via ensureGatewayProbed) handles the path everywhere.
• Probe /api/sessions on the dashboard URL (#276, #289 by @Sanjays2402) — /api/sessions lives on the Hermes Dashboard (default :9119), not the agent. The dev server was hitting the agent URL and getting a 404 every 15 seconds.
• Send Authorization: Bearer $HERMES_API_TOKEN on dev-mode probes (#288, @Interstellar-code) so a gateway secured with API_SERVER_KEY is detected correctly. Same patch added a watch.ignored list for .runtime/.tanstack/.omc/.omx/coverage/dist/etc. so noisy internal paths don't trigger spurious dev reloads.
• Vite proxy port + /cron → /jobs link fixes (#283, @Interstellar-code) — fixes #277 ("/cron 404 Page Not Found").
• Route-tree noise fix — renamed src/routes/api/send-stream-live-tools.ts → -send-stream-live-tools.ts so TanStack Router stops emitting "does not export a Route" on every regeneration.

🛟 Safety

• Update banner shows the full reason and the blocking files (#293) — the banner used to truncate at "...has local c" and never told the user which files were blocking the update. Now displays the full reason, the repo path of the dirty checkout, and a list of up to 8 paths from git status --porcelain. Reason copy mentions "remove the listed files" so users with untracked files know what to do.

🧠 Swarm

• HERMES_TMUX_BIN override (#244) — for non-standard tmux installs (Docker containers, NixOS, custom package layouts). Set HERMES_TMUX_BIN=/path/to/tmux and the swarm dispatcher and runtime will use it instead of falling through the hard-coded candidate list. Also added /usr/bin/tmux to the candidate list (Debian/Ubuntu default).

🔧 Operator surfaces & dashboard

• Provider mix donut + cache efficiency tile, Velocity + Cost Ledger menu entries, and the rebranded dashboard header.
• Contextual Operator Tip card that fills the bottom-left gap; sessions list now stretches to fill the column.
• Polish on swarm surfaces; quieter usage-meter error handling.
• Update flow now realigns clean repos to remote on update.

Breaking / migration

None. This is a strict superset of v2.1.3.

Thanks

Big thanks to community contributors this cycle:

• @Interstellar-code — #279, #281, #282, #287, #288, #283
• @Sanjays2402 — #289 / #276
• @yasuhidekoizumi-afk — #290
• Everyone who filed issues #244, #276, #277, #285, #291, #292, #293

🌙 Onward.

Hermes-Workspace v2.1.3 — pairing, sessions, and branding cleanup

Hermes-Workspace v2.1.3

Hotfix included

• GitHub source, README, docs, config examples, Docker metadata, and release-facing text now consistently use Hermes Workspace / Hermes Agent naming.
• v2.1.3 tag now points at the post-branding-fix commit.

Hermes-Workspace v2.1.3 packages the real post-2.1.2 fixes that landed after the first Swarm release.

Included

• Real Hermes Agent session loading fix for 3002
• Dashboard token compatibility fix so workspace sessions, skills, and jobs use the correct Hermes dashboard auth
• Hermes API, dashboard, and config root alignment
• Visible Hermes Workspace / Hermes Agent branding cleanup across key UI surfaces
• Final browser title, manifest, onboarding, settings, and status-message cleanup

Verification

• Production build passes
• 3002 loads real Hermes sessions
• Core gateway/dashboard capabilities verified live
• No obvious hardcoded secret/private-key leaks found in tracked source, docs, or public assets during audit

Links

• Docs / repo: https://github.com/outsourc-e/hermes-workspace#readme
• Release: https://github.com/outsourc-e/hermes-workspace/releases/tag/v2.1.3

Hermes-Workspace v2.1.x Patch Notes

Hermes-Workspace 2.1 is the first real Swarm release, turning Hermes-Workspace from a single chat surface into a multi-agent control plane for persistent Hermes workers.
Highlights

Swarm, built into Hermes-Workspace

New Swarm orchestration surface inside Hermes-Workspace
Built-in worker coordination, assignment, runtime visibility, and review flow
Persistent multi-agent workflow instead of disposable one-shot subprocesses
Persistent worker runtime

Swarm workers are designed around persistent Hermes sessions
Better tmux-backed runtime behavior for live worker sessions
Cleaner dispatch flow into active worker sessions
Better alignment between worker runtime, task flow, and visible UI state
Reports, routing, and operator flow

Worker dispatch, checkpointing, and review flow improved
Better task routing and reviewer/inbox-style orchestration
Improved runtime and worker status visibility
Better Swarm UI polish around planning, cards, and orchestration surfaces
Hermes naming + runtime cleanup

Cleaned up shipped repo/UI naming so Hermes branding is much more consistent
Follow-up runtime cleanup to align worker/session flow with Hermes Agent
Improved path/config/session handling around Hermes-oriented roots and runtime assumptions
What changed across 2.1

Swarm and Swarm2 UI improvements
better worker controls
cleaner routing and orchestration flow
improved live-session dispatch behavior
better persistent worker semantics
improved HERMES_HOME / ~/.hermes alignment
cleaner runtime metadata and health reporting
reduced hardcoded local path assumptions
added release docs and integration detection groundwork
Architecture direction

v2.1.1 — Swarm runtime pairing fix

Hermes-Workspace v2.1.1

This follow-up release fixes the important runtime pairing issue after v2.1.0.

Highlights:

• Swarm runtime now aligns with regular Hermes Agent worker paths
• Live worker session start/dispatch flow uses Hermes runtime semantics more consistently
• Better HERMES_HOME / ~/.hermes pairing for tmux-backed workers
• Cleaner health/runtime metadata around the active agent path
• Additional cleanup of hardcoded local path assumptions

Verification:

• Production build passes
• Quick repo sweep found no obvious hardcoded secrets or private keys in tracked source/docs/public assets

Recommended if you want the Swarm/Hermes Agent pairing fix included in the release you point people to.

v2.0.0 — Zero-Fork Release

Zero-fork release. Clone, don't fork.

Project Workspace now runs on vanilla pip install hermes-agent (0.10.0+). No patches, no forks, no drift. Upgrade any time with pip install -U hermes-agent.

Highlights

• Zero-fork architecture — workspace talks directly to vanilla hermes-agent via standard endpoints
• One-liner installer — curl -fsSL … | bash provisions everything
• Conductor (/conductor) — mission-control surface: spawn missions, assign workers, watch live output
• Operations (/operations) — agent registry: pause, steer, kill live agents with role + model insight
• Hermes-Nous theme — dark + light editorial variants with cobalt/paper surface pass
• Synthesized tool pills — inline tool-call rendering from dashboard stream markers
• Landing parity pass — hero, features, screenshots, setup, OG image, mobile theme toggle

Fixes

• Duplicate responses and disappearing history on interrupt (#62)
• Portable-mode double user message + uncleaned timeouts
• Local model selection actually propagates to chat
• Dashboard token injection on / instead of /index.html
• Onboarding no longer stacks behind workspace shell

Full notes: CHANGELOG.md

Upgrade

git pull
pnpm install
pnpm build

Pair with latest agent: pip install -U hermes-agent

v1.0.0 — Hermes Workspace 1.0

🚀 Hermes Workspace 1.0

The first official release. Your AI agent's command center — chat, files, memory, skills, and terminal in one place.

Highlights

• Chat — Real-time SSE streaming, multi-session, file attachments, slash commands, voice input
• Files — Workspace file browser with Monaco editor, image preview, file operations
• Terminal — Full PTY terminal with 256-color support
• Memory — Browse, search, and edit agent memory with markdown preview
• Knowledge Browser — Document viewer for agent knowledge bases
• Skills — Browse 2,000+ skills from the marketplace, install/uninstall, security scanning
• Profiles — Multi-profile management (create, switch, rename, delete)
• MCP Settings — Configure MCP servers from the UI
• Jobs — Cron scheduler with create, edit, enable/disable, run history
• Dashboard — Workspace metrics and activity overview
• 8 Themes — Official, Classic, Slate, Mono — each with light and dark variants
• Mobile PWA — Full feature parity on any device via Tailscale
• Security — Auth on all API routes, CSP headers, rate limiting, exec approval prompts
• Smart Routing — Route simple queries to cheaper models automatically
• Context Tracking — Token usage monitoring with alerts

Compatibility

Works with any OpenAI-compatible backend (Ollama, LM Studio, vLLM, OpenAI, Anthropic, etc.). Enhanced features like sessions, memory, skills, and jobs unlock automatically when connected to a Hermes gateway.

Quick Start

git clone https://github.com/outsourc-e/hermes-workspace.git
cd hermes-workspace
pnpm install
cp .env.example .env
pnpm dev

Stats

• 198 files changed, +14,463 / -3,654 lines
• Build: clean | TypeScript: 0 errors | Tests: 4/4 passing