Platform Recovery, Mainnet Bitcoin & Production Guardrails

The last stretch has been less about shiny features and more about making the platform dependable. We had real infrastructure damage, stale Railway state, broken backend wiring, and public surfaces that were showing the wrong thing. This post is the short version of what we fixed.

Railway Recovery

The Agentbot production stack had drifted badly after Railway damage and stale configuration paths. We relinked the project to the correct Railway environment, restored the backend control plane, and got the OpenClaw runtime path back into a working state.

• Reattached production to the correct Railway project
• Restored the backend service and health contract
• Repaired frontend → backend → runtime provisioning flow
• Removed stale host assumptions from active app code paths

Base Wallet First

We made Base the clear default wallet rail. Sign-in, send/receive, and wallet UX are now aligned around the actual Base account path instead of mixing product concepts together.

• Base-first wallet page and receive flow
• Cleaner send state and recent activity
• Sponsored-first USDC send path where available
• Tempo/MPP kept as a secondary agent billing lane

Bitcoin Advanced Rail on Mainnet

Bitcoin is still the advanced rail, not the default onboarding path, but it now runs on mainnet. We rebuilt the NBXplorer path on Railway, cut the backend over to mainnet, and removed the old testnet path from deployment.

• Mainnet `bitcoind` + `bitcoin-backend-mainnet` live on Railway
• Agentbot backend now reads the mainnet NBXplorer service
• Old testnet services and orphaned volumes were cleaned up
• Bitcoin dashboard now better reflects the advanced watch-only wallet path

Queueing and Control-Plane Hardening

We also spent time making the platform less fragile under load and less dependent on lucky client behavior.

• Provisioning moved onto durable backend jobs
• Heavy chat/gateway work was pulled off the hottest request paths
• Worker separation improved between API and background execution
• Guardrails added around concurrency and deployment flows

Production Protection

One of the most important changes was operational, not visual. We started treating the production Railway project as something that needs explicit protection from accidental damage, including from AI-assisted ops work.

• Added a staging environment for safer changes
• Created a scoped Railway project token for production operations
• Rotated live control-plane token usage away from the broader token
• Added a runbook in the repo for future production changes

Why This Matters

We want Agentbot to feel like something you can actually rely on, not just something that demos well. That means repairing the boring parts, tightening drift, and cleaning up production assumptions when they go stale. This is the kind of work that keeps the rest of the roadmap possible.