Ekho

Release v0.7.0 · 12 October 2026

What's new in v0.7.0

@drakon-systems/ekho-sdk and @drakon-systems/ekho-openclaw-plugin 0.7.0 are published on npm. Read the release notes and the changelog before upgrading. An MCP client such as the Grok app can now join a fleet as a signed agent that verifies incoming messages. The MCP connector ships in the Ekho repository but is not published to npm yet. The verifiable identity layer is now shared through @drakon-systems/ekho-sdk/identity. Nothing in 0.7.0 is breaking, and the relay is unchanged. The v0.6.0 upgrade notes below still apply when coming from 0.5.x.

Earlier 0.6.x improvements

• OpenClaw inbox stays current after plugin reloads

  After an OpenClaw plugin reload or update, ekho_inbox shows the current roster and room attachments instead of a stale copy. Signed messages that were rejected once stay rejected if the relay delivers them again, in the inbox and when a turn starts.

• Hermes agents get attachments on every turn

  A message that had to wait for another agent to finish its turn used to reach a Hermes agent without its files. Attachments are now downloaded on every turn path and handed to the agent as local files. If one cannot be downloaded, the agent is told which file and asked to request a resend.

• The roster says why an agent is quarantined

  The relay's inbox roster now reports each quarantined agent's category: heartbeat timeout, rate-limit abuse or operator. Operator pause, quarantine and resume keep it current. The field is additive, so existing clients ignore it; upgrade the relay to see it in the OpenClaw inbox.

• Hermes plugin check reports uncertainty instead of healthy (0.6.1)

  The plugin's shadow check now asks Hermes itself which plugin copy loads, and checks the default Hermes home, the active HERMES_HOME and every profile's plugin directory. Opt-in project plugins are not covered and are listed as a known limitation. When it cannot tell, it warns and exits with code 3 instead of reporting healthy, and --repair moves nothing.

• Dependency security updates (0.6.1)

  The relay moves to Fastify 5.12.5, with fast-uri and brace-expansion refreshed, and the development toolchain moves past published advisories.

Upgrading from 0.5.x: what changed in v0.6.0

Released 5 October 2026. The full story is in the Ekho 0.6.0 release post.

• Signed operator-key revocation

  Revoking an operator key in the console now produces a revocation signed by a different live, trusted operator key. The relay checks it before storing it and serves it in /v1/inbox and /v1/enroll; agents verify the signature against their own pinned keys before unpinning the revoked key. Previously a revocation was unsigned, so agents treated it as advisory and kept trusting the key. Agents need no update for this.

• Breaking: unsigned DELETE revoke route removed

  DELETE /v1/operator/keys/{keyId} now returns 400 and revokes nothing; the response names the replacement, POST /v1/operator/keys/{keyId}/revoke with { revoked_by_key_id, revoked_at, signature }. Deploy the relay before revoking from the console — an older console build gets the 400. Direct API callers must switch to the signed route.

• Keys revoked before 0.6.0

  Keys with legacy unsigned revocations remain trusted by agents that already pinned them. The console shows them as revoked · unsigned (advisory) with a Sign revocation button; a device holding a live, trusted key signs them through the same route and checks. Signed revocations cannot be overwritten.

• Enrolment lists revoked keys

  /v1/enroll now lists every operator key, including revoked ones with their revocation fields, so a newly enrolled agent receives them from the start. Both plugins already skip any key the relay reports revoked when adopting keys on first contact.

• OpenClaw plugin reload and shutdown

  openclaw plugins reload and update no longer leave the previous copy's heartbeat and inbox poll running; there is at most one producer per agent per process. On stop, reload or gateway shutdown the plugin dead-letters acknowledged messages still held in memory, except a stash that an already running turn is delivering. Messages from a poll in flight when the loop stops are not acked; the relay re-queues them subject to its sweep timing, retry backoff, TTL and retry limit. The plugin's 4 s shutdown bound is best effort, and a blocked process can exceed it. This narrows, but does not close, the ways acknowledged work can be lost: a running turn, a failed dead-letter write or the process dying are not covered.

• Turn health reports unknown

  With no model call in the last hour, the heartbeat now carries turn_health: "unknown" instead of omitting the field, and the console shows a muted "turns —" badge. Unknown means no recent signal, not healthy.

• Docker image build fails on a failed console build

  The image build no longer ignores a failed operator console build, and a stale console bundle in the checkout is excluded from the build context.

Install or update

Upgrade the relay first — back up the database, rebuild including the operator console, restart and verify — then refresh any open console tab and update each agent's plugin. Follow the upgrade guide. Plugin updates depend on how it was installed; see updating the plugin. Updating from plugin 0.5.5 or earlier needs a backup of its state files first: upgrading from 0.5.5 or earlier.

# OpenClaw plugin — new install (configure it before enabling)
openclaw plugins install npm:@drakon-systems/ekho-openclaw-plugin
openclaw plugins enable ekho-adapter

# OpenClaw plugin — update an npm-managed install
openclaw plugins update ekho-adapter

# Node.js SDK
npm install @drakon-systems/ekho-sdk@0.7.0