Stacks/Coding agent
Ramp Inspect background coding agent
Ramp Inspect is a company-wide background coding agent built on Modal Sandboxes. Each session gets a full local-like environment (Postgres, Redis, Temporal, RabbitMQ, OpenCode, a browser) that starts in seconds from a snapshot no more than 30 minutes old. The agent is reachable from Slack, a web IDE, and a Chrome extension, and Ramp reports it starts about half of merged PRs.
- Use case
- An internal background coding agent, used by engineers, product managers, and designers at Ramp, that the company says now starts roughly half of all merged pull requests across frontend and backend repos
- Category
- Coding agent
- Team size
- not stated
- Orchestrator / framework
- Each Inspect session runs in a Modal Sandbox with a full-stack development environment. Inside the sandbox, OpenCode is the coding agent, alongside a VS Code server, a web terminal, and a VNC stack with Chromium for visual verification. Outside the sandbox, Modal Functions on a 30-minute cron build fresh filesystem snapshots; Dicts manage session locks and image metadata; Queues route prompts from Slack, the web, or a Chrome extension into the right session. Sessions support multiplayer collaboration and child sessions for parallel work.
- Memory
- Filesystem snapshots, refreshed every 30 minutes (repos cloned, dependencies installed, initial builds run), are the session starting point. Session state is shared across clients so Slack, web, and the Chrome extension stay in sync. A longer-term agent memory product is not described.
- Evals
- not stated as a formal eval suite. Adoption is reported as the impact metric: within a couple of months, roughly half of merged PRs are started by Inspect, and over 80% of Inspect itself is now written by Inspect.
- Observability
- The setup is wired into Ramp's observability stack, including Sentry and Datadog, and into GitHub, Slack, and Buildkite. The original design goal also included deep integration with LaunchDarkly and Temporal so the agent could verify its own work.
- Monthly cost
- not stated
- Models
- not stated
- Other tools
- Modal SandboxesPostgresRedisTemporalRabbitMQVS Code serverVNC and ChromiumGitHubSlackBuildkiteLaunchDarklyChrome extension
- Source
- https://modal.com/blog/how-ramp-built-a-full-context-background-coding-agent-on-modal
Rahul Sengottuvelu and Zach Bruggeman (Ramp), as told in a Modal blog post by Greta Workman · Feb 18, 2026
Related
More like this
Coding agent
Date not stated
Stripe Minions
Unattended end-to-end coding agents at Stripe. Over 1,300 merged pull requests each week are completely minion-produced and human-reviewed, with no human-written code in those PRs
Coding agent
Sep 28, 2026
Replit Agent (model-directed harness)
Replit's production coding agent, whose core loop decides how hard to think, when to hand work off, and which subagent (and model) should do it
Coding agent
May 14, 2026
The coding harness behind GitHub Copilot in VS Code
The agentic coding experience inside Visual Studio Code: assembling context, exposing and executing tools, and running the agent loop that turns model output into editor actions