get started

changelog

What shipped

  1. July 14, 2026

    Keys are claimed, not emailed

    • Your API key is never sent by email. After checkout, /welcome shows it once — or use the single-use claim link from your inbox. A compromised inbox later holds a dead link, not a credential.
    • The billing webhook is now credential-free: it provisions your org and a claim token; the key is minted the moment you claim it, and only once.

  2. July 14, 2026

    Built for agents

    • A machine-readable quickstart at looptail.ai/docs/agents — served as raw markdown at /docs/agents.md — written for AI agents: exact commands to record, verify, evaluate, and improve, no account required.
    • AGENTS.md in the repo, refreshed llms.txt / llms-full.txt, and explicit AI-crawler welcomes in robots.txt.
    • A Claude Code plugin: /plugin marketplace add maxfain/looptail installs a skill that instruments any project with looptail.

  3. July 12, 2026

    The full loop in TypeScript

    • @looptail/sdk 0.3.0 brings the entire loop to JS/TS — Observe, Understand, and Improve — at full parity with the Python SDK: instrumentAnthropic / instrumentOpenAI, evaluators (run), failure clustering (cluster), and the Improve loop (propose / replay / canary / approve). A JS verdict and a Python verdict agree.
    • The CLI gains looptail evals run, issues cluster, and improve.
    • Provider SDKs are optional peer deps, so the core stays zero-dependency.

  4. July 12, 2026

    Self-serve plans

    • Start a Team or Scale plan without waiting on us: pick a plan, pay through Stripe, and your API key lands in your inbox — the org is provisioned automatically.
    • Cancel anytime; your plan downgrades cleanly and your exported archive is always yours.

  5. July 12, 2026

    Your dashboard, hosted

    • A customer dashboard at ingest.looptail.ai/app: pick an app and see pass-rates by rubric, tracked issues, chain verification, and anchors — all from your synced trail, all org-scoped to your key.
    • New read API behind it: GET /v1/apps (apps overview) and GET /v1/issues/{app} (clustered issues, latest per key).

  6. July 12, 2026

    Hosted evaluation: the loop runs on our side

    • Upload a rubric to the hosted Tail and the platform judges your synced decisions continuously — no cron of your own, no eval code to run.
    • Every verdict is Ed25519-signed by the server key, so it is independently verifiable, like a receipt — and never touches your own signed chain.
    • New API: POST /v1/rubrics, GET /v1/rubrics/{app}, GET /v1/evaluations/{app} (verdicts + pass-rate). Same rubric evaluator as looptail-evals, so hosted and local verdicts agree.

  7. July 12, 2026

    Security release 0.2.1

    • Cross-language hash fix: canonical JSON now sorts object keys by Unicode code point in every implementation, so a trail written in Python verifies byte-for-byte in JS even with astral-plane keys. Update to SDK/CLI 0.2.1.
    • Ingest hardening: 1 MiB body caps, per-IP rate limiting on public endpoints, and a strict CORS allow-list.
    • The site now runs Astro 7; npm audit reports zero vulnerabilities across the monorepo.

  8. July 12, 2026

    The Improve loop closes

    • looptail-issues — failing verdicts cluster into tracked issues by rubric and weakest criterion. Deterministic, explainable, cumulative.
    • looptail-improve — propose (LLM patch drafts from recorded failures), replay (your regression set, judged by rubric), canary, and approve. Every step a signed improve/approve event; canary and approve are gated on a passing replay.
    • verify --anchors — the CLI now checks the server’s signed chain-head receipts against your local trail: receipt signatures verify, and every anchored head must exist locally.
    • Public-beta plumbing: self-serve API key issuance and monthly plan limits with a soft-cap grace band.

  9. July 12, 2026

    Evaluators, instrumentation, and the trail viewer

    • looptail-evals — score recorded trails against rubrics you write, with an LLM judge on models you choose; verdicts land in the trail as signed evaluate events.
    • Auto-instrumentation for the OpenAI and Anthropic Python SDKs: wrap the client once, every call leaves a trail.
    • The hosted trail viewer at ingest.looptail.ai — paste a key, see your trail, receipts included.
    • The full 10-post launch library: the open Tail format announcement, EU AI Act deep dives, eval-driven development, incident postmortems, judge-cost math, and more — plus LangSmith and Braintrust comparisons.

  10. July 11, 2026

    Signup opens; evidence export ships

    • looptail.ai/signup — request a private-beta invite (and every button on this site now goes somewhere real).
    • looptail export — a dated, self-verifying evidence pack: full chain, manifest, and instructions the recipient can run without trusting you.
    • The founder runbook: deploys, publishing, and key issuance, documented end to end.

  11. July 11, 2026

    The hosted Tail begins

    • Ingest service, code complete: server-side re-verification of every event, append-only storage, and signed receipts anchoring each chain head.
    • SDK & CLI reference — the whole API surface on one docs page.
    • Architecture and development docs published in the repo.

  12. July 11, 2026

    The SDKs are real

    • looptail (Python) and @looptail/sdk (TypeScript): local-first, hash-chained, Ed25519-signed trails.
    • npx @looptail/cli verify — check any trail from your terminal.
    • The open trail format spec, draft v0.1, published in the repo.
    • Trails are cross-verifiable: a trail written in Python verifies with the JS CLI.

  13. July 11, 2026

    looptail.ai is live

    • The brand, this site, and the docs quickstart.
    • The EU AI Act record-keeping guide.
    • Private beta onboarding for design partners begins.