Releases

Changelog

What's new in Pipes.Run desktop + cloud. New releases land here within minutes of shipping.

v0.1.1 — 30 June 2026

The marketplace lands as one unified surface: free public discovery and paid datasets share the same model, the same listing form, the same URL, and the same buy flow. Lower-friction for publishers, simpler mental model for buyers.

Marketplace

  • One marketplace, free + paid. /marketplace is now a single index that shows every listed dataset with a clear price chip ("Free" / "$49" / "$19/mo"). Setting a price on the listing form flips a listing paid; clearing it makes it free again. No separate storefront URL.
  • Paid listings check out via Stripe — one-time charges or recurring monthly/yearly subscriptions. Inline price_data means no pre-created Stripe Price per dataset to maintain.
  • Listing detail page (/d/<slug>) renders the same template for both modes. Free → row preview + open download. Paid + no purchase → description, schema, table sizes, license terms + Buy CTA, with row values hidden until purchase. Paid + valid mp_access cookie → full preview + download. Public dataset slug stays the canonical URL — bookmarks survive a listing's price changes.
  • Access cookie (mp_access, HttpOnly, SameSite=Lax, 30 days) is dropped when a buyer follows the receipt link. Subsequent visits to the dataset page see the gate open without re-auth.
  • Subscription lifecycle is wired: customer.subscription.deleted / canceled / unpaid flips the matching DatasetPurchase to canceled, revoking access at the next download attempt.
  • Listing form gains optional price_cents, billing_interval (one_time / month / year), and license_terms fields. Validation rejects sub-$0.50 (Stripe floor) and >$9999.

SEO / public surfaces

  • /robots.txt + /sitemap.xml ship — allow public, disallow dashboard/admin/auth/api/buy/checkout/billing. Sitemap only lists listed datasets (rung 3 of the privacy ladder), honoring the publisher's choice not to be indexed.
  • OG/Twitter meta tags on every /d/<slug> so dataset shares to Slack, Twitter, LinkedIn, Discord render with title + hook.
  • ProxyFix middleware (x_proto=1, x_host=1) so _external=True URLs produce https://pipes.run/... behind Cloud Run's TLS- terminating frontend. Affects OG canonicals, sitemap entries, and magic-link emails.

Marketing site

  • Marketplace surfaced in landing-page Product footer + the shared footer on every other page. (Header promotion deferred until ~10 listings exist.)
  • Landing-page FAQ entry explaining how the public marketplace coexists with the local-first promise (opt-in publishing, two confirmed clicks before anything is visible).
  • Privacy §2.4 + Terms §11 added covering marketplace publishing: what becomes public, what we deliberately don't expose, UGC license buyers grant us to host, publisher warranties, DMCA takedown flow, no-warranty disclaimer on third-party datasets. "Last updated" bump to 30 June 2026. Banners still mark these as draft pending counsel review.

Reliability

  • dataset_published_at TIMESTAMP column added to cloud_pipelines (was DATETIME in an earlier rev — Postgres has no DATETIME type, caused a boot crash that's now fixed).
  • view_count actually increments on each /d/<slug> view (atomic SQL UPDATE, not read-modify-write). The model promised it; now the route delivers.
  • Removed duplicate dataset_public + dataset_slug column definitions from CloudPipeline.

v0.1.0 — 25 May 2026

The first public release. Everything below ships in v0.1.0; later releases will be incremental updates from here.

Desktop app

  • Chat with the web, save as pipeline. Three search verticals (web, news, past day/week/month/year), DuckDuckGo + Bing + Brave with automatic fallback. Local LLM query rewriting turns chat-style prompts ("find AI funding announcements from this week") into clean keyword queries before hitting the search engine.
  • 10 LLM providers. Anthropic, OpenAI, Gemini natively; Perplexity, Groq, Mistral, Together, OpenRouter, Cohere via LiteLLM; Ollama for local models with auto-detection.
  • Multi-step transforms with {row} / {row.field} / {field} placeholders so each step sees the prior outputs.
  • Three destinations: SQLite (default), Postgres, DuckDB.
  • Write modes: append, replace, upsert (with compound dedupe keys).
  • Schedule in plain English ("every weekday at 9am") or cron.
  • Sources: web search, crawl, HTTP/JSON, CSV, Excel, Google Sheets (OAuth), SQL query against any of your databases.
  • Workspace backup — every pipeline + provider as one YAML file. API keys are intentionally excluded.
  • Run cancellation, failure notifications, pipeline dependencies, run history modal, conversation history drawer all in.
  • First-run health check detects missing crawler + unconfigured LLM providers and offers one-click fixes.
  • Auto-update banner + license validation with 7-day offline grace.
  • Signed + notarized macOS DMG — no Gatekeeper warnings, installs like any Mac app.

Cloud

  • Marketing site (landing, pricing, privacy, terms) with hero chat mockup, FAQ, comparison table.
  • Magic-link sign-in via Cloudflare Email Worker + MailChannels.
  • Stripe Checkout for Desktop ($99 one-time) + Cloud subscription ($19/month). Webhook-verified license minting; idempotent retries.
  • Customer dashboard with license, download, cloud-pipeline list, recent runs.
  • Desktop ↔ cloud sync API (/api/v1/pipelines, /api/v1/runs) for pushing pipelines to always-on cloud infrastructure.
  • Mock cloud executor runs a background thread that picks pending runs and processes them — to be replaced with the real pipes.runner.run_pipeline for real cloud execution before launch.

Admin / ops

  • Admin user view (/admin) for support: search users, see their licenses + seats, reset seats, comp a license, revoke a license, resend the receipt email. Every action requires a reason field that gets persisted to the audit log.
  • Audit log of every admin action — actor, target, action, reason, timestamp.