Skip to content
Start free

Changelog

What you can do now,and since when.

LiteSurface is in early access: Solo and Team are free while we build in the open. Each release is written as what changed for you; the technical notes are folded underneath.

Follow with RSS
  1. Latest

    A redesigned workspace and admin console

    The product app and the admin console now share one design language, so the same controls look and behave the same everywhere.

    • You can now move between projects, concepts, and evaluations in a workspace rebuilt on a shared design system.
    • Admins now get an operations dashboard for providers, prompts, jobs, usage, and audit in one place.
    Technical details
    • Shared console design system used by both the user app and the separate admin app.
    • Admin console redesigned as an operations dashboard; admin UI stays in its own app.

  2. Merge concepts, link contradictions, duplicate projects

    Evolving a portfolio got easier: combine overlapping ideas, see where evidence disagrees, and start a new thesis from a proven setup.

    • You can now merge two concepts into one and keep the lineage of both.
    • You can now link claims that contradict each other, so disagreement in the evidence stays visible.
    • You can now duplicate a project and start the next thesis from the same setup.
    • You now get reminders when experiments are due, capabilities go stale, or new evidence contradicts a concept.
    Technical details
    • Concept merge with a lineage tree; contradiction links between claims.
    • Project duplication, user jobs, and reusable recipes.
    • Reminder digests by email for due experiments, stale capabilities, and contradicting evidence.

  3. Scores now wait for enough evidence

    A confident recommendation now needs enough cited evidence behind it. When coverage is thin, LiteSurface researches what is missing before it recommends.

    • You can now trust that thin evidence caps the recommendation instead of hiding behind a high score.
    • Concepts below the evidence minimum get a research pass aimed at what is missing, then a fresh evaluation.
    • You can now run research against the live web with Tavily or Brave search.
    Technical details
    • Evidence coverage became a soft gate (35% minimum by default) with an evidence-gap loop (ADR-015).
    • Claims are triaged for decision relevance at extraction time; retrieval ranks by it.
    • Live research verified with Tavily; SSRF range fix, provider-text fallback, parallel extraction.
    • Concepts generate in parallel chunks and fail fast on truncated model output.

  4. Budgets, an admin console, and whole-project export

    Teams can now run LiteSurface with guardrails: spending limits that actually stop runs, provider and prompt controls, and an audit trail.

    • You can now set monthly soft and hard budgets per workspace; a hard limit refuses new runs instead of overspending.
    • Admins can now check provider health, reset a circuit breaker, and roll prompts back to an earlier version.
    • You can now export a whole project as a bundle.
    Technical details
    • Separate admin app: provider health with breaker reset, prompt versions with runtime overrides and rollback, jobs with retry, usage dashboard, audit.
    • Hard budget limits answer HTTP 402. Usage ledger per project and run.
    • Deterministic eval suite (schema compliance, citation integrity, injection resistance, evaluation consistency) as a CI gate.
    • Accessibility sweep with axe; Content Security Policy for production builds; OpenAPI 3.1; backup and restore drill; runbooks.

  5. Generate a build-ready handoff

    A concept that earns its place can now leave LiteSurface as a package a coding agent can execute.

    • You can now generate a validation brief, a PRD, or a 26-section implementation handoff for any concept.
    • You can now regenerate a single section without touching the rest, and every citation resolves to a stored claim.
    • You can now download the bundle as a ZIP, pinned to the concept version it was built from.
    Technical details
    • Blueprints with section-by-section generation and a citation guard.
    • Completeness and consistency validators plus an AI review; Markdown bundle with footnoted SOURCES and DECISIONS.
    • ZIP export with expiring, authenticated downloads; per-section regeneration creates a new revision.

  6. See every score decomposed, then test the riskiest assumption

    Scores became something you can take apart, and the next step after a score became a plan rather than a guess.

    • You can now see each criterion with both evaluators’ scores, the weight, the median, and the gates in one view.
    • You can now compare concepts side by side and keep a scorecard per project.
    • You can now plan experiments with kill criteria, record outcomes, and re-evaluate without rewriting the earlier result.
    Technical details
    • Evaluation UI: score decomposition, hard gates, compare view, project scorecards.
    • Validation engine: ranked assumptions, experiment planning, experiments board, structured outcomes, non-mutating re-evaluation.

  7. Generate, version, and fork concepts

    Ideas became structured concept genomes that can be compared, evolved, and decided on, scored by OpenAI and Anthropic independently.

    • You can now turn opportunities into concepts with a typed genome: user, job, trigger, mechanism, wedge, model, and assumptions.
    • You can now version and fork a concept, and record decisions with a rationale next to the AI recommendation.
    • Near-duplicates are flagged before they clutter the portfolio.
    Technical details
    • OpenAI Responses and Anthropic Messages adapters with structured outputs, repair, retry, fallback, and pricing.
    • Opportunity discovery, concept generation with diversity constraints, dedupe judge, versions, decisions, forks.

  8. Workspaces, the capability atlas, and cited evidence

    The foundation: projects you own, a map of what platforms can do, and evidence stored with a locator for every claim.

    • You can now create workspaces and projects, invite people with roles, and share or transfer a project with everything in it.
    • You can now build a capability atlas and add sources that are broken into atomic, cited claims.
    • You can now watch long research runs progress live and resume after a failure.
    Technical details
    • Turborepo monorepo, BetterAuth workspaces, project-centric ownership with transfers, audit and activity.
    • Capability atlas, SSRF-safe research fetcher, claim extraction with pgvector retrieval, durable runs with SSE progress.

Get the next entry by emailinstead of checking back.

Product updates, about once a month. Unsubscribe in one click. Privacy

Try what shippedon an idea of your own.

Solo and Team are free during early access. Explore a sample project before you run your own.

Start freeSee a sample evaluation
  • Solo and Team are free during early access
  • Explore a sample project first
  • Bring your own OpenAI and Anthropic keys