Hyperbach

Data feeds you can build on: jobs, ads, video transcripts, social profiles, products, reviews and search data, as clean, typed rows. We build for scheduled pipelines, internal tools and AI agents as much as for people.

Pay per result. A row charged is a row delivered; a blocked page, a retry or an empty run costs nothing.

Monitor mode on every Actor. A scheduled task with it on delivers and charges only what changed.

No login, no account risk. No Actor asks for your account, cookies or session.

Honest limits. Every README names the walls and what the Actor does at each.

Flat rows, typed schema. Every dataset declares its schema; rows need no mapping.

Reliability. 568,884 runs in the last 30 days, 99.9% of them succeeded; 803,548 since 2025. 540 acceptance scenarios on the deployed build before and after every release; 11,412 automated tests.

Monitoring and incidents. Every run reports status, rows, timing and the source's health to our telemetry. Actors, sources and proxies are checked every 15 minutes and a change alerts us at once. A fix ships as a build: 261 builds in the 7 days to 2 November 2026; every Actor's live build is from today.

Architecture. One shared, tested kit under every Actor: retries, per-row charging, monitoring memory and telemetry are built once and reach the fleet in one release.

Interface. Every Actor takes JSON and returns a typed dataset, through the API or Apify's MCP server. Every Actor carries a real example input and a row cap, so a run's cost is known before it starts, the same per row at any volume; charges never exceed rows, 261 acceptance checks at every release. A run ends with a status line: what was delivered and why it stopped. 56 of 57 READMEs carry an agent section with the prompt.

Experience. Built and run by an engineer with 30 years in software.

Questions, requests, a source you need: apify@hyperbach.com