Olympus is a read-only HermesOS Agent Monitor for operational tuning.

It reads local Hermes runtime state and shows what to tune next across profiles, routes, skills, sessions, cron, gateways, Kanban work, and runtime health. It does not run agents and does not replace Hermes admin pages. It explains how the agent system is performing and links to the Hermes page that owns each fix.

See OLYMPUS_GOAL.md for the product boundary (what Olympus owns vs. what Hermes owns).

Olympus dashboard showing readiness score, mode tabs, and Agent Monitor

• Ranks agent optimization items with the evidence behind each one.
• Compares profile readiness, route metadata, skill coverage, gateway state, and workload.
• Summarizes Kanban pressure: blocked work, active workers, retries, and assignee load.
• Surfaces safe config policy risks: turn limits, loop guardrails, browser privacy flags, fallbacks, toolsets, and auxiliary cost visibility.
• Computes a transparent heuristic readiness score with a full deduction breakdown.
• Opens with a Brief view for the next action, then stages deeper panels behind Agents, Skills, Kanban, Policy, and Diagnostics tabs.
• Keeps v1 read-only: every action is a link to the Hermes page that owns it.

• Create or move Kanban tasks, or spawn workers.
• Edit profiles, cron, gateways, routes, memory, MCP, keys, or config.
• Rebuild Hermes Usage/Analytics for token, model, session, or spend totals.
• Show secrets, or expose local labels/paths unless explicitly enabled.

• A working HermesOS / Hermes Agent install ($HERMES_HOME, default ~/.hermes).
• Python 3.10+ (the backend is a FastAPI router loaded by Hermes).
• Node.js/npm for verification and Playwright-based visual smoke tests. There is no frontend build step; dashboard/dist/* is hand-authored SDK React/CSS.

Install Olympus from source, link it into your local Hermes home, and start the Hermes dashboard:

git clone https://github.com/AIEngineerX/Olympus.git
cd Olympus
npm ci
npm run dev

npm run dev links this repository's dashboard/ folder into:

$HERMES_HOME/plugins/olympus/dashboard/

Then it starts Hermes with:

scripts/install-dashboard-link.sh
hermes dashboard --no-open --skip-build

Then open:

http://127.0.0.1:9119/olympus

After the tab loads, follow after-install.md to confirm the first screen, privacy defaults, and local smoke checks.

Olympus opens in Brief mode: hero status, score details, and Agent Monitor. Deeper inspection is split into purpose-built tabs:

• Agents: Performance Tracking, Profile Fitness, and Pantheon.
• Skills: Skill Coverage and Skill Hygiene.
• Kanban: Trace Spine and Kanban Intelligence.
• Policy: Tool Policy & Aux Cost.
• Diagnostics: Operational Evals, Production Diagnostics, and Evidence Sources.

Hermes Desktop owns chat, agents, profiles, skills, sessions, cron, settings, and Command Center Usage. Olympus is the read-only operations monitor beside those controls. It should show risk and evidence, then link to the Hermes page that owns the fix.

Today Olympus is verified through the Hermes web dashboard at /olympus. Desktop visibility requires Hermes Desktop plugin-tab parity for dashboard plugins. See dashboard/docs/HERMES_DESKTOP_INTEGRATION.md.

Run npm run test:desktop to verify the local Desktop app, Command Center Usage ownership, web dashboard plugin support, and the current plugin-tab parity status before preparing an upstream Desktop PR.

Hermes mounts Olympus at /api/plugins/olympus/:

Routes sit behind the Hermes dashboard session-token middleware. The frontend calls them through the plugin SDK's fetchJSON, which injects the token.

Olympus hides session titles, Kanban task titles, cron names, exact route labels, and local paths by default. To show richer local labels on a private machine:

OLYMPUS_EXPOSE_LOCAL_LABELS=1 hermes dashboard --no-open --skip-build

See SECURITY.md for the full security and privacy posture.

.
├── dashboard/
│ ├── manifest.json # plugin metadata, tab path, assets, api module
│ ├── plugin_api.py # FastAPI router (read-only backend)
│ ├── dist/index.js # hand-authored SDK React frontend
│ ├── dist/style.css # olympus-* styles
│ ├── docs/BUILD_PLAN.md # implementation plan
│ ├── docs/HERMES_DESKTOP_INTEGRATION.md
│ ├── docs/PRODUCTION_READINESS.md
│ ├── docs/VIEWPORT_STRATEGY.md
│ ├── docs/FRONTEND_SKILL_RESEARCH.md
│ └── README.md # package notes
├── docs/assets/olympus-dashboard.jpg
├── scripts/install-dashboard-link.sh
├── scripts/olympus-live-smoke.mjs
├── scripts/olympus-performance-smoke.mjs
├── tests/visual/ # Playwright fixture checks
├── tests/fixtures/ # Olympus browser fixtures
├── package.json # verification scripts
├── after-install.md # post-install checks
├── OLYMPUS_GOAL.md # product boundary
├── TODO.md # active backlog
├── SECURITY.md # security and privacy notes
├── CONTRIBUTING.md # development and contribution guide
├── LICENSE # MIT
└── README.md

Note: dashboard/dist/* is hand-authored SDK React/CSS. There is no bundler or build step. See CONTRIBUTING.md for the development workflow.

npm run verify
npm run test:visual
npm run test:live
npm run test:performance
npm run test:security
npm run test:desktop
npm audit --audit-level=moderate

npm run test:visual uses a fixture-backed Playwright harness. It loads the hand-authored dashboard assets directly, checks desktop/mobile readability across multiple fixture states, validates link destinations, verifies empty evidence sections stay hidden, verifies staged dashboard modes, and catches private-label leaks in the no-labels scenario.

npm run test:live checks the real Hermes dashboard route at http://127.0.0.1:9119/olympus. It starts Hermes when needed, verifies desktop/mobile rendering, clicks each dashboard mode, and fails on console errors, overflow, tiny labels, bad links, SVG text, or missing staged panels.

npm run test:performance checks the real /overview and /tuning plugin API routes through the Hermes session-token flow. It fails on non-2xx responses, slow local response time, backend diagnostic budget warnings, or payloads above the documented budget.

The live smoke runner reuses an existing Hermes dashboard only when the served HTML includes the Hermes session token. If it needs to start Hermes and your $HERMES_HOME/plugins/olympus/dashboard target already points somewhere else, set OLYMPUS_SMOKE_RELINK=1 before running the smoke command.

• API responses redact session IDs, private labels, paths, and secrets unless OLYMPUS_EXPOSE_LOCAL_LABELS=1 is explicitly set.
• Brief mode hides deep panels by default; visual and live smoke tests click the tabs that own Agents, Skills, Kanban, Policy, and Diagnostics.
• Pantheon uses HTML text and accessible profile buttons rather than SVG text or an image-role wrapper.
• Frontend refreshes ignore stale /overview responses so an older request cannot overwrite newer data.
• Attention items are sorted by severity before truncation.
• Gateway configuration detection parses .env variable names only; values are not scanned or returned.
• Config policy reads only safe structure, counts, and flags; it does not return prompt text, base URLs, API keys, env values, or local paths.
• Dependencies are exact-pinned in package.json, locked in package-lock.json, and checked with npm audit --audit-level=moderate.

Contributions are welcome. Please read CONTRIBUTING.md first. It covers the local dev loop, the read-only and privacy rules, the verification commands, and commit/PR conventions.

Found a security or privacy issue? See SECURITY.md for the posture and how to report.

• HermesOS / Hermes Agent
• HermesOS documentation
• dashboard/docs/PRODUCTION_READINESS.md
• dashboard/docs/VIEWPORT_STRATEGY.md

MIT, AIEngineerX.