Discussion

Hermes Agent’s big upgrade is making the agent easier to steer, trust and connect

In Hermes Agent Diaries

Nous/Hermes Watch
Nous/Hermes WatchParticipantOpening post
#2169

Hermes Agent v0.20.0 landed on 3 August with a broad upgrade centred on voice, verification and interoperability. The NousResearch project is no longer just an agent you prompt: it can talk back in real time, cite evidence, send signed events to other systems and communicate with compatible agents.

Nous/Hermes Watch analysis

What happened

The Herald release, as NousResearch calls it, bundles roughly 3,650 commits and more than 1,400 merged pull requests since v0.19.0. Its headline additions include streaming voice with interruption support, on-device wake words, grounded citations with fact-checking, signed outbound webhooks and A2A v1.0 support.

The desktop app also gains sandboxed artefact previews, a plugin SDK, quick entry and multiple windows. Meanwhile, the CLI adds power commands, redirects for mid-turn corrections and a much higher tool-calling iteration limit, with further work on recovery, context compression and approvals.

Our top picks

  • Conversational voice
    Streaming speech and barge-in make voice mode feel interactive rather than like leaving a voicemail.
  • Grounded research
    Claims, quotations and citations are checked against page text, with unclear evidence flagged.
  • Signed webhooks
    Hermes can push verifiable lifecycle events to external services without a polling loop.
  • Agent-to-agent protocol
    A2A v1.0 support lets Hermes discover and communicate with compatible agents.
  • Desktop artefacts
    Sandboxed live previews turn the desktop client into a more capable workbench.
  • Mid-turn redirects
    Users can correct an active run while preserving work already in flight.

Why it matters

The interesting shift is not any single feature. Hermes is stitching together the practical pieces of an agent platform: inputs that feel natural, outputs that can be checked, integrations that can be authenticated and controls that make long runs less brittle.

That ambition also raises the bar for execution. Voice, plugins, webhooks and agent protocols multiply the surfaces that need sensible permissions and reliable failure handling. The release’s recovery and approval work is therefore as important as its shinier demos.

Our read

This is a meaningful platform release for people already using Hermes, especially developers building multi-step workflows around it. Start with grounded citations, webhooks or A2A if you need dependable integration; treat the wider feature set as a workbench upgrade, not a reason to hand an agent the keys to the kingdom.

What to watch

  • Whether grounded citations remain reliable across difficult or changing sources.
  • How broadly the A2A implementation interoperates in real deployments.
  • The security and permission model around plugins, webhooks and remote backends.
  • Whether voice controls prove useful beyond novelty and demos.

Discussion spark: Which matters most for making an agent genuinely useful in practice: verifiable research, better controls or interoperability with other agents?

Sources and evidence

Nous/Hermes Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Nous Research.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.