v0.1.0 Release Notes

v0.1.0 is the first public release of Hermes Agent Pod. It turns the sandbox concept into a reproducible local runtime kit for nousresearch/hermes-agent:latest, with both kind and Docker Compose paths, a localhost dashboard, and a Codex-oriented worker wrapper.

For a workflow-focused tour, read the v0.1.0 walkthrough.

Highlights

• Added a local Hermes runtime that can start as a sandbox-hermes kind Pod or as the sandbox-hermes-agent Docker Compose service.
• Exposed the Hermes gateway on 127.0.0.1:8642 and the dashboard on 127.0.0.1:9119, with guidance for LAN-only dashboard viewing while keeping the gateway local.
• Added scripts/hermes-worker for bounded Codex delegation through the OpenAI-compatible /v1/chat/completions endpoint.
• Switched the shipped runtime defaults to Z.AI GLM Coding Plan: provider zai, model glm-5.1, and base URL https://api.z.ai/api/coding/paas/v4.
• Published bilingual VitePress documentation and a GitHub Pages deployment workflow.

Runtime Kit

The release includes compose.yaml, kind-config.yaml, and Kubernetes resources under k8s/. The Kubernetes path seeds /opt/data/SOUL.md from prompts/hermes-worker-system.md, writes runtime configuration from a ConfigMap, and stores agent state in the hermes-data PVC. The Compose path uses data/ as the host-mounted runtime state directory.

Both runtimes are intentionally local-first. The gateway and dashboard bind to localhost by default, and the docs call out that external exposure needs a reverse proxy, authentication, and network controls.

Worker Delegation

scripts/hermes-worker.py provides the Codex-facing wrapper. It accepts a prompt from arguments or stdin, can attach text files with --file, supports session continuation through --session-id, and lets operators override the base URL, API key, model, timeout, and system prompt.

The wrapper keeps the contract narrow: Hermes receives bounded context through the API, while Codex remains responsible for host-side file changes and final decisions.

GLM Defaults And Secret Handling

scripts/set-glm-key.sh configures the GLM key for the Compose runtime and patches the kind Secret when a kind-sandbox-hermes cluster is available. The release also keeps real keys out of Git with ignored local files and placeholder-only Secret examples.

The runtime defaults are implemented in compose.yaml, k8s/hermes-pod.yaml, and scripts/up.sh, so both Compose and kind paths share the same provider, model, base URL, and longer timeout expectations.

Docs And Release Surface

The documentation site ships with English and Japanese guides for setup, usage, architecture, and troubleshooting. The release also adds a dashboard screenshot asset and GitHub Actions workflow for publishing the VitePress site to GitHub Pages.

Validation

Before publishing this release note set, the release work checked the tag history with collect-release-context.ps1, validated SVG assets with verify-svg-assets.ps1, and built the VitePress docs with npm run docs:build.

Runtime smoke tests that need Docker, kind, or live API keys should be run in the target operator environment before relying on the gateway for production-like delegation.