About
We build a decision systemfor the choice before the code.
LiteSurface exists to help teams choose what to build with the same rigor they bring to building it: evidence first, scores you can take apart, and a record that outlasts the meeting where the call was made.
Mission
The most expensive decisionis usually made on a hunch.
Teams spend months building and minutes deciding what to build. The choice is often made from a deck, a strong opinion, or whichever idea was described most vividly. When it turns out wrong, the reasoning is rarely recoverable.
AI has made ideas cheap. A model will produce fifty concepts before lunch and describe each one convincingly. That makes choosing harder, not easier, because fluency is not the same as evidence, and a chat transcript is not a record.
Our mission is to make the choice as rigorous as the build: every concept structured, every score traceable to its sources, every decision kept with its reasons.
Why a decision system
Not another assistant.A place where decisions live.
An assistant answers and moves on. A decision system keeps the objects a decision is made of, links them together, and lets you come back to them.
A conversation
- One model’s opinion, phrased persuasively
- Sources mentioned in passing, if at all
- A different answer each time you ask
- Gone when the thread scrolls away
A decision system
- Two independent evaluators on a published scorecard
- Every claim linked to a source and a locator
- Deterministic scores you can recompute from stored results
- Durable, versioned objects with lineage and history
Principles
Six commitmentsthat shape the product.
Evidence before eloquence
A persuasive paragraph is not evidence. Claims carry their source and a locator, coverage measures how much of a judgement rests on them, and thin evidence becomes a research task rather than a confident guess.
Models propose, code keeps the score
Models read, draft, and judge. Deterministic code computes medians, weights, gates, and confidence, so any stored result can be recomputed exactly and explained line by line.
Disagreement is information
Two evaluators from different providers score every criterion independently. Where they differ, we show it, because an average that hides an argument is worse than no number at all.
History is never rewritten
Evaluations are immutable. New evidence, experiment outcomes, and human overrides are appended beside what came before, so you can always see how a decision was reached.
Your work stays yours
Projects are the unit of ownership. You bring your own model keys, choose which providers may be used, and can export or transfer a project with everything in it.
Honest about limits
A score is a structured judgement, not a forecast or a legal opinion. Partial results are labeled as partial, and anything still on the roadmap is described as planned.
What we will not doeven when it would be easier.
Principles are only useful if they rule something out. These are the shortcuts we have decided not to take.
- Resell model tokens or mark up your provider spend
- Show a score without the evidence and rules behind it
- Average away disagreement between evaluators
- Edit an evaluation to match a decision made later
- Present an AI score as legal, safety, or financial approval
Where we are todayearly access, in the open.
LiteSurface is in early access. The pipeline runs end to end today, from the capability atlas to the build handoff, and a sample project shows a complete run before you connect a provider. Some capabilities, such as SSO and hosted data residency, are still on the roadmap and are labeled that way wherever they appear.
Put the principlesto work on your next idea.
Explore a sample project before you run your own, then connect OpenAI and Anthropic when you want live research and evaluation.
- Solo and Team are free during early access
- Explore a sample project first
- Bring your own OpenAI and Anthropic keys