Ramp built its own Laminar agent harness to stay model- and provider-agnostic

omarsar0 · x · 2026-10-23

Sakana AI fan account? No—this is omarsar0 highlighting Ramp's engineering blog post on Laminar, the in-house agent harness that now powers all of Ramp's agentic systems: an accounting harness, a general coworker called glass, a coding agent, the Ask Ramp support bot, headless OCR agents, and Ramp Text. It supports any model plus modern harness features: background tasks, sandboxes, tool hooks, tool search, durability, forking, compaction, queueing, streaming, and observability.

Why not use managed agents APIs? Ramp wanted model- and provider-agnosticism to optimize cost, latency, and performance, and found wrapping multiple vendors' evolving APIs roughly as much work as building their own. Building alongside product teams also means minutes-from-request-to-shipped feedback loops no external vendor can match. The payoff: faster shipping, in-house agent expertise, and full ownership of the observability stack.

Original post →

More from coding & agent

coding & agent channel →