TOOL PROFILE / EDITION 2026.09
Hermes Agent + Desktop
A self-improving AI agent from Nous Research. It is public and MIT licensed, works with multiple model providers, and ships CLI, TUI, and Desktop surfaces with persistent memory, skills, subagents, MCP support, terminal backends, and browser automati
No ordinal score is published because comparable evidence is incomplete. No rank delta is shown because there is no prior numeric snapshot.
Best for
Persistent, delegated work across projects you own end to end
Price & access
Free and open source (MIT). Costs are your own model API usage and hosting.
Evidence summary
The published rubric scores visual fidelity, iteration, deployment, and value against frontend-specific evidence. For Hermes Agent that evidence is not first-party or independent yet, so inventing subscores would fabricate precision. It is tracked, searchable, and contextualised next to coding agents until the evidence exists.
Edition status
New. September 5, 2026: desktop app and OpenRouter 30-day global #1 usage status independently surfaced in the index.
Strengths
- MIT licensed and fully self-hostable, so nothing is locked to a vendor
- Persistent memory, skills, and subagents give it continuity other agents lack
- Multi-model: you choose the engine underneath
- MCP, terminal backends, and browser automation in one runtime
Tradeoffs
- Frontend-specific visual quality is not first-party benchmarked
- Not a visual builder — there is no design canvas or preview-first workflow
- Setup, hosting, and model spend are your responsibility
- Evidence for the site's seven frontend dimensions does not yet exist
Linked evidence
OpenRouter apps marketplace ↗OpenRouter Hermes profile ↗Hermes Desktop ↗Hermes Agent repository (Nous Research) ↗Hermes Agent product site ↗Use the same published data, filters, and side-by-side metrics.