<- all tokdocs

Ramp Laminar: How Ramp Runs Agent Sandboxes and Credentials

By , founder of Mavric and Apso

View on TikTok ->

Laminar is Ramp's in-house managed agents platform, described on the Ramp Builders Blog on October 20, 2026. A TikTok from @jakestechjourney recommends Ramp's engineering blog and lists its agent system as an orchestrator, a pool of sandboxes, access control, least privilege and a credential broker. The advice to read the blog is good, but the post that matches that list published the day after the video, and it rejects a sandbox pool by name.

Where is Ramp's engineering blog?

The video says to "look up Ramp's engineering blog." Ramp's publication is the Ramp Builders Blog at builders.ramp.com, and engineering.ramp.com serves the same posts. The archive runs from "Elixir at Ramp" on May 24, 2021 to "Agentic Credit Analysis" on October 22, 2026.

Its agent posts, in order:

  • "Building Ramp's MCP server" (March 25, 2025)
  • "How To Build Agents Users Can Trust" (July 10, 2025)
  • "Meet Ramp Research: Our Agentic Data Analyst" (September 18, 2025)
  • "Why We Built Our Own Background Agent" (January 12, 2026)
  • "We proactively fixed ~100 security issues in 6 days with 0 humans" (February 20, 2026)
  • "Agentic identity: modeling agents to keep users in control" (May 4, 2026)
  • "Agentic Risk Operations" (June 30, 2026)
  • "How we built OCA, our AI on-call assistant" (September 9, 2026)
  • "Building Laminar: Ramp's Managed Agents Platform" (October 20, 2026)

The video posted on October 19, 2026. The Laminar post, by Sam Kronick, is the only one that covers the architecture the video describes. None of the earlier posts use "orchestrator," "credential" and "broker" together, and the January background agent post uses none of those words.

What is Laminar?

The Laminar post says Ramp tried "just about every 'agent harness' or platform out there," including model-provider SDKs, managed agent APIs and open source coding agents, then built its own for provider independence. "Around May 2026 we decided to take what we learned and build towards a common platform." "The v1 of Laminar was built in 6 weeks," and "Five months later, every agentic product at Ramp runs on Laminar."

It runs on what the post calls boring technology: "Postgres, Temporal, and S3 for persistence and workflow execution." Its parts include a tool catalog, durable tool dispatch, sandboxes, subagents and human approvals.

The post also says "Laminar came into existence soon after the first managed agents APIs launched." That undercuts the video's line that Ramp was "ahead of their time."

Does Laminar use a pool of sandboxes?

No. Kronick writes: "from day 1 we made sure the entire stack runs on a fleet of generic disposable workers ('cattle') rather than a pool of carefully tended sandboxes or servers ('pets')."

Ramp calls the pattern sandbox-as-a-tool, as opposed to agent-in-a-sandbox. The sandbox deep dive says "instead of maintaining a sandbox for every active session, we can run many agent loops on shared workers and offload code execution to isolated compute environments as needed." A session's history and state live in a database, so "losing a sandbox doesn't necessarily mean losing the session." Sessions "are granted access to sandboxes rather than owning them," and a parent agent can share one sandbox with its subagents.

A sandbox pool does appear in the earlier background agent post, about Ramp's coding agent Inspect. There it is a speed optimization: "You can also keep a pool of warm sandboxes for any high-volume repositories," expired and recreated as new images build. Those sandboxes run on Modal from per-repository images rebuilt every 30 minutes. That post reports "~30% of all pull requests merged to our frontend and backend repos are written by Inspect."

How does Ramp handle agent credentials?

Ramp never writes "credential broker." It describes two components that do that job.

The egress proxy handles network calls. "Sandbox traffic goes through our egress proxy, which enforces destination restrictions outside the sandbox and can attach credentials to outbound requests on the agent's behalf." An agent can fetch a private package without holding the secret, so it cannot leak into a generated script or "the output of an enthusiastic printenv." Ramp sums it up: "This separates permission to use a credential from possession of its value."

The Connectors Service handles integrations. It "centralizes authentication, credential management, access controls, and tool execution behind one API," and "controls who can use each credential and what they're allowed to do with it."

The phrase "least privilege" does not appear in the Laminar posts. Ramp's only use of it is in the March 2025 MCP server post: "following the principle of least privilege is as crucial as ever."

Is there an orchestrator in Laminar?

Not as a named component. The post uses the word once, defining a harness as "the thing that orchestrates and manages that loop." The function is spread across Temporal workflows in Laminar.

For older work on scoped access, Ramp's October 11, 2023 post on just-in-time access describes moving from three AWS SSO roles to more than twenty for "fine-grained access control."

Who made the video

The account belongs to Jake Hulberg, whose site lists him as "Developer advocate @ Infisical," a company that sells secrets management and privileged access tools. His site describes a PlatformCon talk on "Agent Vault, an open source credential broker," and his workshop repo was created on 2026-06-21. "Credential broker" is his own term mapped onto Ramp's design. The video does not mention this. I found no published video of his rebuild of Ramp's system as of 2026-10-26.

Bottom line

Ramp's blog is worth reading, and the Laminar post is the one to start with. Its design runs agent loops on disposable shared workers, treats sandboxes as tools a session borrows, and keeps credentials outside the sandbox in an egress proxy and a connectors service.

Sources

Published October 19, 2026. Based on a TikTok by @jakestechjourney. Full video notes (markdown)