Sandbar

Agentic surf forecasting that reads a break like a local — translating buoy data and NOAA grid forecasts into a spot-specific call on when to paddle out.

Problem

Surf forecasting is broken for people who actually surf. Generic swell forecasts report open-ocean numbers — wave height, period, direction — but can't account for how local bathymetry, wind timing, and tide phase interact at a specific break. What reads as a solid 4-foot swell at the buoy might arrive as closeout slop or a perfect A-frame depending on dozens of variables the forecast ignores.

The result is forecasts that are technically accurate and practically useless. Surfers learn to distrust them and develop their own mental models from experience — which works, but only for breaks they know well and only if they have the time to triangulate multiple sources by hand.

Insight

The forecast signal lives in the combination of sources, not any single one. Buoy data is ground truth — it tells you what the ocean is actually doing right now. NOAA grid forecasts are context — they tell you what's coming and from where. The ensemble logic that reconciles them against historical spot behavior is the differentiator — it's where ambiguous signal becomes something legible and trustworthy enough to act on.

Good surfers already do this synthesis intuitively. They check the buoy, glance at the swell chart, factor in the tide, and make a call based on how all of it has played out at that break before. Sandbar makes that process systematic — and available to anyone, for any break, without years of local experience.

Product

Sandbar's core output isn't a number — it's a timeline. For a given break it predicts how surf quality rises and falls across the morning and marks the window actually worth paddling out for. What a generic forecast flattens into "4ft all day" becomes a shape: fair at dawn, building to a short best window, then declining as the tide or wind turns.

PREDICTED SURF QUALITY ACROSS THE MORNING — THE WINDOW WORTH PADDLING OUT FOR

Because the right wave depends on what you ride, every call is board-aware — the same morning can read "good" on a log and "poor" on a shortboard. You describe your quiver once, and board compatibility factors into the score from then on. The question Sandbar answers isn't "what are the conditions?" — it's "when should I paddle out?"

Under the Hood

The agentic pipeline ingests buoy observations and NOAA grid forecasts on a rolling schedule, reconciles them against historical spot behavior, then scores the output for confidence. Low-confidence forecasts are flagged rather than hidden — if the model isn't sure, the user should know that.

BUOY DATA → NOAA GRID → ENSEMBLE MODEL → SURF FORECAST

That score is a composite. Separate modules evaluate swell alignment, wind, tide, and learned break behavior — each contributing to a single surf-quality score, then weighted by how well the conditions suit the board you're on. It's the synthesis an experienced local does intuitively, made explicit and repeatable.

SWELL · WIND · TIDE · BREAK BEHAVIOR → ONE SCORE, TUNED TO YOUR BOARD

The ensemble adapts its source weighting to conditions — leaning on buoy data when swell is long-period and well-organized, weighting NOAA grids more heavily when a system is still offshore. The goal is to replicate the judgment a local surfer applies without requiring the years of observation it normally demands.

What's Next

Real-time buoy webhooks will replace the current polling schedule — bringing latency down from minutes to seconds when a swell event arrives. Coverage is expanding to more breaks, with each new spot improving the ensemble's generalization. The longer-term goal is a community validation loop that lets local surfers correct the model and sharpen it over time — turning accumulated local knowledge into a shared, machine-readable resource.

Sandbar is in private beta. If you surf and want early access, reach out.