Tabstack
Mozilla's "Web Execution Layer" API: Extracts web page content, autonomous research, and browser automation into a single API call. Input URL + schema to get structured JSON, input a question to get an answer with citations, or input natural language tasks for a hosted browser agent to complete multi-step workflows. Core automation engine Pilo is open source, reads the accessibility tree rather than screenshots, and official benchmarks claim it uses 60-80% fewer tokens than screenshot-based approaches. Models, browsers, and orchestration all run on Tabstack; developers do not need to set up their own headless browser clusters.
Mozilla-backed Web Execution Layer: One call delivers structured results, accessibility-tree engine saves 60-80% tokens, data is discarded immediately, no model training, default compliance with robots.txt.
Who it is for: Developers and teams building AI agents / research tools / data pipelines
Core capabilities
- /extract/json + /extract/markdown: Input URL and custom schema to get consistently matched structured JSON; or convert any page to clean Markdown directly for LLM context. Schema enforcement, rendering, and parsing are all completed within the call.
- /generate/json: Perform inferential structured output based on custom instructions, not just field extraction, can transform and generate from page content.
- /research(Autonomous Research): Complete source discovery, reading, synthesis, and citation in one call, stream responses with inline citations, answers come from real-time web pages not pre-indexed cache; available in Fast / Balanced modes.
- /automate(Browser Automation): Describe tasks in natural language, hosted browser agent navigates, clicks, fills forms, and completes multi-step workflows on sites you don't control; supports interactive mode to pause and ask humans for context when encountering login or missing context, guardrails limit executable actions.
- Pilo Engine(Open Source): Self-developed execution engine driving /automate, based on Playwright accessibility tree for Observe-Decide-Act agentic loop, includes layered retries, context compression (verbose tag mapping + deduplication, saving 60-80% tokens), secondary LLM quality check; MIT-style open source, fully self-hosted with API key (github.com/mozilla/pilo).
Pricing
Freemium(10,000 Free Credits + Pay-as-you-go / Subscription)
Updated 2026-08-17