System
What is running, how it is configured, and whether it answers.
Does it answer?
Three separate things can be broken, and they look identical from a failing page. These tell them apart.
The agent
What answers a tenant, and whether a human sees it first.
Auto-approve is on: proposals can execute without anyone seeing them.
Providers this build can construct a client for: none, fake, ollama, openai, anthropic, openrouter, groq.
Settings
Every setting the engine resolved at boot, grouped by what it affects. Secrets are reported as set or unset — their values never leave the server.
Where records live. Account approvals always need Postgres — they sit outside the store on purpose, so they are the one feature that fails alone when DATABASE_URL is missing.
| OWNAWAY_STORE_BACKEND | json |
| OWNAWAY_STORE | data/store.json |
| DATABASE_URL | not set |
| OWNAWAY_DB_DRIVER | neon |
| DB_POOL_MAX | 5 |
Which model answers, which agent version runs it, and whether it may act without review. OWNAWAY_SEND_MODE=suggest is what keeps a human in front of every message.
| OWNAWAY_MODEL | groq |
| OWNAWAY_AGENT | v5 |
| OWNAWAY_SEND_MODE | set |
| OWNAWAY_AUTO_APPROVE | true |
Deployment
Four Workers on one Cloudflare account, each with a staging twin. The commands are the same ones a deploy actually runs.
Cloudflare account e9911ce467e2896b1c8a7e90c78a8b6c