What Linear actually runs

Linear’s stack has a reputation for being exotic. It isn’t — and the company says so itself. CTO Tuomas Artman, describing the architecture: “We chose a ‘very normal’ tech stack, based on TypeScript. We have React on the frontend and Node on the backend… We use Postgres as a database, and Redis for queuing… and GraphQL as our API layer. And all of it runs on GCP with Kubernetes.” The exotic part is a single deliberate exception: a homegrown real-time sync engine, built in the company’s earliest days, that everything else is arranged around.

Sourcing note: our baseline is Artman’s 2022 deep-dive with The Pragmatic Engineer; current state comes from the “Our tech” section Linear publishes on its own careers page (accessed July 2026). Where the two lists disagree, that’s a migration signal — labeled inferred, not confirmed. Full record in the stack database.

Stack at a glance

Context: ~30 employees at the 2022 interview (15 engineers) serving what the careers page now describes as 25,000+ companies including OpenAI, Coinbase, and Ramp (confirmed). Third-party aggregators report ~178 employees (Latka) and an $82M Series C at a $1.25B valuation in June 2025 (Sacra) — both reported, not company-confirmed.

The one exotic thing, built first

Most startups bolt real-time sync on later. Linear inverted that. Artman: “In the very early days of Linear, we built a complex engine which took care of synchronizing the frontend and the backend, with offline support… Because we did this so early on, it meant that as we grew, engineers could focus on building features… as they would not have to think about networking, error management, race conditions or data synchronization between clients.” (Pragmatic Engineer, confirmed)

The mechanics, per third-party reverse-engineering of the client (inferred, 2022-vintage — details may have evolved): the app keeps a near-full local copy of workspace data in IndexedDB, bootstraps via a sync endpoint, then streams live changes over WebSockets as ordered sync actions, replaying anything missed after a reconnect. Postgres sits behind it as the system of record. Notably, analysis of the engine suggests Linear’s collaboration model relies on a centralized server establishing canonical ordering rather than the CRDT approach favored by much of the local-first movement (inferred). MobX provides the reactive object graph on the client — the piece that makes a locally-synced database feel like an instant UI. Artman’s own 2023 talk, “Scaling the Linear Sync Engine”, covers how it held up (confirmed it exists; the details live in the video).

Infrastructure decisions made like migrations are expensive

Two founding-era choices show the same philosophy: pay upfront to never migrate. Kubernetes arrived when the entire product team was two engineers and a designer — “sounds like bad prioritization,” Artman concedes, but “sooner or later, we’d need to move over to Kubernetes, so we might as well bite the bullet now… it means we won’t have to migrate” (confirmed). It’s the mirror image of Cursor’s Yugabyte retreat — where Cursor bought scale-readiness it didn’t need and walked it back, Linear pre-paid for infrastructure it grew into.

The cloud choice was less philosophical. On GCP versus AWS: “in my view GCP is moving ahead on cost-for-performance. In the end, we went with GCP because the user interface is much better than AWS” (confirmed). Sometimes the deciding factor for a billion-dollar company’s cloud is that the console is nicer.

What changed between 2022 and now

Comparing Artman’s 2022 list against Linear’s current careers page yields two deltas, both unannounced:

MongoDB is gone. “In place for caching scenarios” in 2022; absent from the company’s own current stack list (inferred — could be careers-page shorthand rather than a true retirement; we’ve asked for verification). Temporal arrived, appearing twice in the current list — once in the core backend line-up and once powering an “agentic workload system… at production scale” (confirmed). The timing lines up with Linear’s 2025 push into AI agents, and their public GitHub org now includes Cursor and Claude agent integrations (confirmed). Durable workflow orchestration showing up exactly when a fast-moving product adds long-running AI tasks is the most 2026 stack move there is.

What we couldn’t confirm

In keeping with our method: we found no reliable public evidence for Linear’s auth provider, billing stack, or observability tooling, so they’re not in the record — rather than guessed at. If you work at Linear, the record is one pull request away from a ✓ verified badge.

What we’d watch

Whether the sync engine’s centralized-ordering model holds as Linear pushes deeper into multiplayer AI-agent territory, what actually replaced MongoDB’s caching role, and whether “very normal” survives the agentic era — Temporal is the first crack in the boring-on-purpose doctrine, and we mean that as a compliment.

Sources: The Pragmatic Engineer deep-dive with CTO Tuomas Artman (baseline), Linear’s careers page “Our tech” (current state), marknotfound, wzhudev, Sacra. Research assisted by AI agents, verified and edited by a human. Confidence labels: confirmed / inferred / reported. Spot an error? We correct publicly.