Deploy and Host Hippocamp Dream with Railway

Run Hippocamp's memory compaction overnight without keeping your computer awake. This template creates a scheduled Railway service that clones your private Lagoon memory repository, compacts projects over the configured wake-up threshold, and opens one reviewable pull request per project.

About Hosting Hippocamp Dream

The service runs once per day at 03:17 UTC and exits after it finishes. Each run starts from a fresh clone, writes only current_state.md and open_threads.md, and leaves append-only events untouched. GitHub remains the audit trail and review surface; Railway only hosts the scheduled job.

Required Configuration

• LAGOON_REPOSITORY: your private memory repository in owner/repository form
• GITHUB_TOKEN: a fine-grained GitHub token limited to that repository, with Contents and Pull requests read/write access
• MANIFEST_BASE_URL: a Manifest or OpenAI-compatible Responses API base URL
• MANIFEST_API_KEY: the API key for Dream model requests

Common Use Cases

• Compact agent memory while your laptop is offline
• Keep long-lived project wake-up context concise
• Review every automated memory rewrite through a pull request
• Run the same Dream workflow across multiple projects in one Lagoon repository

Dependencies for Hippocamp Dream Hosting

Deployment Dependencies

• A private GitHub repository containing Hippocamp memory
• A fine-grained GitHub token scoped to that repository
• A Manifest or OpenAI-compatible Responses API endpoint and key

Why Deploy Hippocamp Dream on Railway?

Railway keeps the scheduled compaction job available while your computer is offline. The service is ephemeral, runs on a daily cron, and connects directly to the open-source Hippocamp repository so template deployments can receive upstream updates.

The local Hippocamp MCP remains local-first and does not depend on Railway.