Why We Built Our Own Background Agent

Summary

Ramp built Inspect, a hosted background coding agent that writes code and verifies its work using the context and tools available to engineers. For backend tasks it can run tests, inspect telemetry, and query feature flags, while frontend work can be checked visually with screenshots and live previews. Each session runs in a sandboxed Modal VM initialized from repository-specific images and snapshots, with integrations including GitHub, Slack, Sentry, Datadog, LaunchDarkly, Braintrust, and Buildkite. Inspect supports frontier models, MCPs, custom tools, and skills, and provides Slack, web, Chrome extension, pull request, and browser-based VS Code workflows whose changes remain synchronized. Ramp reports that about 30% of merged pull requests in its frontend and backend repositories are written by Inspect.

Context

Ramp wanted a background coding agent with enough context and tooling to verify its work like an engineer, without relying on a developer’s laptop, local checkout, or worktree. The system was also intended to support multiple concurrent sessions and workflows across engineering and non-engineering users.

Approach / What changed

Use repository-specific images built every 30 minutes, snapshot them after setup, and launch fast Modal Sandboxes from those snapshots. Run an agent such as OpenCode inside each sandbox, connect it to company tools and workflows, and expose sessions through Slack, the web, Chrome, pull requests, voice, and a hosted VS Code editor while synchronizing all changes.

Takeaways

• Repository images are cloned, provisioned, and snapshotted every 30 minutes; new sessions start from the latest stored snapshot, limiting repository staleness to at most 30 minutes.
• The post recommends OpenCode because it is server-first, has a fully typed SDK, and provides a comprehensive plugin system for custom clients and behavior such as blocking edits until synchronization finishes.
• Ramp says Inspect writes about 30% of merged pull requests across its frontend and backend repositories, reaching that level within a couple of months without forcing adoption.

Related reading