Skip to content
Documentation navigation +
implemented

Changelog

The API is in beta (1.0.0-beta). Breaking changes are listed here until v1 is declared stable.

2026-10-17 — Careers browsing

The recruiting hostname is now careers.alivehealthtech.com. Browse internships with keyword and discipline filters, then view the selected role's details, glyph animation and Apply button beside the list (stacked on mobile). Removed the application deadline and introductory tag. Added drawn research and communications glyphs; the email application flow remains unchanged. Custom-domain availability depends on hosting/DNS configuration.

2026-10-17 — Internship applications by email

Five approved internship listings and an application form prepare a prefilled email to team@alivehealthtech.com. Applicants attach a résumé/CV and send it from their email app. The upload control is disabled; copyable application text provides a fallback. The main footer links to the selected Apply host and /careers redirects there. Supabase-backed storage, staff review dashboard, automated emails and online assessments remain pending. This entry describes repository behavior; deployment/custom-domain acceptance is tracked separately.

2026-10-17

  • Authoring sessions (beta): POST /v1/projects/{project_id}/authoring-sessions, GET /v1/projects/{project_id}/authoring-sessions, GET /v1/authoring-sessions/{session_id} and POST .../answers, .../revise, .../approve; CLI alive author start|show|answer|approve; Python SDK start_authoring, authoring_session, answer_authoring, approve_authoring; web page /projects/{id}/author. Additive; no existing contract changed.
  • Authoring checkpoints now include world.settings (each configured value with its unit and source) and world.variants[].changes (the values a variant changes). Additive.
  • Authoring checkpoints' ai object now includes estimated and session_total usage (calls, tokens, AI iterations, wall seconds) and cap_reached when a session stops at its preparation limit. Preparation limits now apply to the whole session, not each revision. Additive.
  • Authoring checkpoints now include world.measured: for each result, its coverage status (covered, partial or missing), unit, denominator, time window, censoring and any measurement reported beside it (for example what is still waiting at the end). Additive.
  • Authoring checkpoints now include next_steps: what to do next, each with kind, an optional Alive tool and source (rule or ai_hypothesis). Additive.
  • New CLI exit code 4: input or approval needed (an authoring session is waiting for answers). Scripts that treated any nonzero code as failure should handle it.

2026-10-13

  • One-time organization/program credit checkout now has a disabled-by-default Stripe candidate: immutable intents, strict paid-session checks, once-only ledger fulfillment and operator reconciliation. Conditional app checkout exposes approved configured prices. Real provider acceptance, paid subscriptions, refund/dispute policy and activation remain pending.

  • Focused Python SDK, CLI, real local SDK/API parity and sponsor/payment contract checks added. CI now checks freshly built Python clients outside a repository checkout. Publication and broader IDE/platform acceptance remain separate gates.

  • M12 begins with an operator-gated manual sponsor pilot: draft programs, activation/closure, enrollment decisions, funded allocations, fixed aggregate reports and a conditional app workspace. Default-disabled; pilot approval/acceptance and paid plans remain pending.

  • Sponsor read-policy recursion and duplicate allocation counters are corrected. Reports separate held reservations from settled use; scheduled programs enforce their start date.

  • Idempotent project/intake/run/review creates now commit work and replay records in one serialized transaction. Changed bodies are rejected for reused keys, including existing pre-checkpoint runs/reviews. Distinct intake/world revision saves serialize revision numbering.

  • M11 adds the thin Python SDK with generated TypedDict annotations, a buildable ESM TypeScript distribution, and shared version-matched guides/schemas/catalog/sample worlds. Package registry publication remains pending. CLI discovery, trace pages, comparisons and resumable errors are documented in the agent guide.

  • Schema discovery no longer requires repository paths. Worker completion preserves canceled state; an already-running in-process calculation is not forcibly interrupted.

  • The terminal study now uses numbered headers and transitions from parallel worlds to three simulated coding-agent sessions. These are illustrations, not live integrations.

  • The completed-slice audit adds a read-only /v1/experiments/compare endpoint for stored baseline/treatment comparisons within one run. Router-level 404/405 errors now use the API error envelope. The error reference includes all current server error codes.

  • The terminal study opens into parallel-world and coding-agent panes, each with its own transcript scrolling. Multiverse outcome-linking is a planned concept; the Claude Code-style conversation is simulated. Wide, intermediate and mobile layouts keep the panes readable.

  • Terminal playback has a startup pause, slower blue developer input, response pauses and file/ background-test acknowledgements. Transcript scrolling unlocks after startup; Follow latest resumes following when reading earlier output. The session remains a workflow preview.

  • Section five uses a smaller blue cursor trail. Desktop navigation is centered and uses the label Methods. Product-page globe art is circular, with clearer world-model explanations.

2026-10-12

  • A new Product page explains applications, the evaluation workflow and current model scope; Client work now explains custom software and scoped integration services.

  • Decorative glyphs retain their bending, with related-character substitutions, heavier local strokes and a fading cursor trail. The terminal preview types developer input and focuses on multiple tests, feedback, iteration and an agent handoff, with smaller glyph worlds and faster pacing. Its workflow calls remain proposed interfaces rather than executable documentation.

  • Homepage navigation now distinguishes keyboard focus from mouse focus when hiding on scroll, including after hash navigation. The temporary backer-logo layout experiment was removed.

  • Careers internships now include concrete simulation and research skill expectations. Forms require a résumé/CV file and allow multiple optional links; drafts include the file locally, with application submission still unimplemented.

  • The homepage terminal previews a future world-function workflow, including tiny glyph worlds, varied output pacing, pause/replay and an expandable transcript. Its proposed functions are illustrative, not current CLI commands. Page scrolling is no longer caught by the terminal.

  • The hero navigation links have matching sizes; localized glyph slant is stronger for visual review. The temporary backer-logo study was subsequently removed.

  • Homepage model details are centered; model artwork varies in color and orientation. The product preview now pairs the dashboard with an expanded scrollable CLI session. The homepage header hides on downward scroll and returns on upward scroll. Section-five worlds have a dedicated finer-detail SVG while small logos retain their original density.

  • Careers lists Winter 2027 internships (November 15, 2026 deadline), two technical full-time roles and advisory opportunities. Custom forms prepare downloadable drafts only; submission and the staff recruiting portal remain planned. See application drafts.

  • Decorative artwork now responds through individual glyphs near the cursor, rather than tilting the whole image. Two-world logos use transparent SVGs with rounded strokes and softened planet edges; updated light and dark SVG downloads are available on the site.

  • Public site and app share a new palette and system-default light/dark appearance with a single toggle. The homepage includes decorative glyph motion, sequential software/CLI design previews and a model explorer; preview projects are illustrative and no command runs there. Developer docs retain static reading surfaces and a compact mobile navigation. See appearance and motion settings.

  • App WorldSpec import accepts JSON files or pasted JSON with compiler validation, a change table, review acceptance before saving, and separate estimate/run approval. The advanced copy prompt includes the public model catalog. Malformed JSON cannot reach the graph preview.

  • Optional WorldSpec cost_scenarios price registered event counts with sourced rate assumptions, low/base/high sensitivity and separate additional costs. Reports are in the API, app and ZIP; unknown costs leave totals undefined. The results endpoint now publishes a typed response.

  • EvaluationPlan and ConceptReview now include an optional business_assessment question map. It distinguishes submitted claims, hypotheses and unknowns; no viability score is emitted.

  • First public contracts: IntakeSpec, EvaluationPlan, WorldSpec, ResourceSpec (experimental), ConceptReview, RunManifest, Finding and MetricSummary 1.0.0.

  • HTTP API /v1 (beta): intake extraction and drafting, intake revisions, evaluation plans and approvals, concept reviews, worlds, runs, events, exports, credits, API tokens, provider keys, research consent.

  • alive CLI: local validation, planning and runs; remote commands with API tokens.

  • Tool registry 2026-10-12.2: concept and evidence review, and synthetic simulation, are available.