Project

AltClaw — one-click AI agent deployments on isolated VMs

Backend / full-stack engineer

Self-service deployment platform for AI assistant agents. Each agent — OpenClaw, PicoClaw, Hermes, or Ottie — runs in its own GCP Compute Engine VM and is paired to Telegram (plus a web dashboard for OpenClaw). Lives at claw.altllm.ai.

Owned the security hardening of the auth, session, and proxy layers; built and stabilised the Playwright end-to-end suite from scratch; and shipped CI + harness infrastructure plus the SSE log viewer users watch while their agent VM warms up.

• Hardened auth, session, and proxy paths to fail-closed by default — eliminated query-string token transport, required explicit JWT secrets in shared environments, anchored session renewal to portal state, and refused requests on portal outages.
• Built the AltClaw Playwright end-to-end suite from scratch — Portal SSO + Google OAuth, deployment lifecycle for every agent variant, paid OpenClaw renewal, team workspace, billing, mobile, and a live-staging synthetic smoke.
• Aligned local validation, harness targets, and runbooks against one source of truth, and added type-aware linting plus coverage thresholds to the CI gate so quality regressions fail loud.
• Shipped a constrained terminal-style SSE log viewer with idle-timeout handling and reconnect logic, plus clearer warming + deployment states across the dashboard.

Gallery