felipe/hermes-agent-custom
1
0
Fork 0
mirror of https://github.com/domfelipe/hermes-agent-custom.git synced 2026-08-07 05:16:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
Hermes Agent custom image with Ollama Cloud config for Railway
33 commits 4 branches 0 tags 109 KiB
Find a file
Felipe Domingues 5d9a8deb16
Some checks failed
quality / quality (push) Failing after 37s
Details
ci: add Forgejo quality workflow (mirror-friendly) 
Runs on self-hosted Forgejo Actions; ignored by GitHub Actions.
GitHub remains source of truth.
2026-08-06 22:02:39 -03:00
.forgejo/workflows ci: add Forgejo quality workflow (mirror-friendly) 2026-08-06 22:02:39 -03:00
.github/workflows feat: post Mika cron and skill actions back to platform (#5) 2026-05-30 14:05:17 -03:00
patches feat: post Mika cron and skill actions back to platform (#5) 2026-05-30 14:05:17 -03:00
plugins/mika_runtime feat: post Mika cron and skill actions back to platform (#5) 2026-05-30 14:05:17 -03:00
tests fix: keep Hermes home aligned in runtime image (#6) 2026-05-30 18:47:27 -03:00
.gitignore feat: post Mika cron and skill actions back to platform (#5) 2026-05-30 14:05:17 -03:00
Dockerfile fix: keep Hermes home aligned in runtime image (#6) 2026-05-30 18:47:27 -03:00
entrypoint.sh fix: keep Hermes home aligned in runtime image (#6) 2026-05-30 18:47:27 -03:00
README.md feat: post Mika cron and skill actions back to platform (#5) 2026-05-30 14:05:17 -03:00

README.md

Hermes Agent - Ollama Cloud Custom Image

Custom Docker image for deploying Hermes Agent on Railway with Ollama Cloud provider and Gemma 4 31B model.

Why this exists

The Hermes gateway ignores environment variables for the primary model configuration and only reads from config.yaml. This image bakes the correct config at build time, solving the fallback to anthropic/claude-opus-4.6 issue.

Configuration

The image pre-configures:

  • Provider: ollama-cloud
  • Model: gemma4:31b-cloud
  • Config path: /opt/data/.hermes/config.yaml

Environment Variables (set in Railway)

HERMES_HOME=/opt/data/.hermes
API_SERVER_ENABLED=true
API_SERVER_KEY=your-secure-key
GATEWAY_ALLOW_ALL_USERS=false
HERMES_SOUL_OVERRIDE=Your custom soul prompt here
HERMES_STT_PROVIDER=local
HERMES_TTS_PROVIDER=disabled
HERMES_GATEWAY_ENABLED=auto
OLLAMA_API_KEY=your-ollama-api-key
PORT=8642
TELEGRAM_BOT_TOKEN=your-bot-token
TELEGRAM_ALLOWED_USERS=your-user-id
TELEGRAM_HOME_CHANNEL=your-channel-id
MIKA_AGENT_INSTANCE_ID=agent-instance-uuid
MIKA_PLATFORM_FUNCTIONS_BASE_URL=https://<project>.supabase.co/functions/v1
MIKA_CREATE_CRONJOB_URL=https://<project>.supabase.co/functions/v1/create-cronjob-from-agent
MIKA_CREATE_SKILL_URL=https://<project>.supabase.co/functions/v1/create-skill-from-agent
MIKA_INTERNAL_FUNCTION_SECRET=shared-internal-secret

MIKA_CREATE_CRONJOB_URL and MIKA_CREATE_SKILL_URL are optional when MIKA_PLATFORM_FUNCTIONS_BASE_URL or SUPABASE_URL is present. The Mika platform provisions all of these automatically for managed Railway services.

HERMES_GATEWAY_ENABLED=auto starts hermes gateway run whenever TELEGRAM_BOT_TOKEN is present, so the container consumes Telegram polling in addition to serving the dashboard/proxy runtime API.

By default the entrypoint hides API_SERVER_KEY from the gateway subprocess so Hermes does not start its native api_server adapter on the same public port as the Mika runtime proxy. Set HERMES_GATEWAY_API_SERVER_ENABLED=true only if you intentionally want the native adapter as a separate gateway platform.

Validation

python3 -m unittest discover -s tests -v
python3 -m compileall plugins patches tests
git diff --check

Deploy to Railway

  1. Connect this GitHub repo to Railway
  2. Railway auto-detects the Dockerfile
  3. Add the environment variables above
  4. Deploy

License

MIT