Tabstack
Tabstack is a Mozilla-built web API for AI applications. It returns Markdown or structured JSON from pages, produces research answers with citations, and automates tasks on public websites. Hosted Automate cannot log in to authenticated sites; Pilo is the separate self-hosted browser automation path.
Agent builders often need the live web but do not want to maintain scraping, browser, proxy, parsing, and LLM orchestration infrastructure. Tabstack matters because it packages extraction, research, and browser automation behind one API while documenting privacy, robots.txt, pricing, and comparison boundaries.
Mozilla New Products describes Tabstack as a web execution and data transformation API for AI systems that need to read and take action on the web. Tabstack docs document the API surface, automate endpoint, credit-based pricing, trust pages, and first-party comparison pages. The open-source mozilla/pilo repository shows the related natural-language browser automation layer, while Hacker News discussion supplies demand evidence around browser infrastructure for agents.
- Give an AI agent structured web data without writing a scraper.
- Run multi-step browser workflows without hosting a browser grid.
- Turn web pages into Markdown or schema-shaped JSON for RAG and enrichment.
- Build competitor, pricing, research, or monitoring agents with cited web evidence.
Tabstack combines several web-access jobs that agent builders often wire together separately: structured extraction, Markdown extraction, generated structured output, cited research, and browser-like automation. The public docs frame it as an API for AI agent builders rather than a general website testing framework.
- Tabstack product and endpoint overview
The publisher distinguishes research, extraction and public-site automation.
- AI Model API: distinguish the model from its web tools
Tabstack supplies web results to an application; a model API defines model inputs, outputs and provider limits.
- Extraction: fetch pages and return Markdown or schema-shaped JSON.
- Research: ask a question and receive a cited multi-source answer.
- Automation: describe a task and stream browser-like progress events while Tabstack navigates, clicks, fills forms, or asks for human input when needed.
Tabstack hosted Automate accepts a task and starting URL, then streams events while interacting with a public website. The official guide checked October 27, 2026 explicitly excludes login and authenticated portals. Interactive mode can request form values; it does not remove that hosted access limit. Pilo is the self-hosted engine for automation in a browser you control. An agent harness using either option still needs task permissions and checks of the returned result.
- Hosted Automate limits and controls
Check the public-site boundary, streaming events, guardrails and interactive input behavior in the official guide.
- Agent Harness: permissions and result verification
See the surrounding runtime responsibilities that a web tool API does not replace.
- Pilo self-hosted repository
Inspect the separate engine and its browser-control documentation before selecting a deployment path.
Tabstack pricing is credit-based across endpoints. The docs list per-action costs, plan tiers, overage behavior, and the difference between pricing and rate limits. The public product copy also emphasizes data handling, robots.txt compliance, and no model training on customer data. Treat anti-detection and heavily protected-site behavior carefully because Tabstack comparison docs say those capabilities are not publicly documented as a Browserbase-style strength.
- Tabstack pricing and plans
Check current credits per action, plan eligibility, overages and spending controls before estimating a workflow cost.
Tabstack is a web execution layer that a harness can expose to an agent with guardrails.
Tabstack is not a model API, but it helps agents gather structured context before model calls.
Browser automation and web access require clear permissions, data handling, and action limits.
Source confidence
Mozilla New Products
Tabstack
Tabstack Docs
Tabstack Docs
GitHub / Mozilla
Hacker News
Tabstack FAQ
Common questions about Tabstack.
Is Tabstack a scraper, a browser automation tool, or a research API?+
Tabstack spans all three jobs. It can extract Markdown or JSON from pages, generate structured output from web content, run cited research, and automate multi-step browser-like tasks through `/automate`. The useful distinction is that Tabstack packages these jobs behind an API instead of asking the developer to host the browser, LLM prompts, and parsing pipeline.
How is Tabstack different from Browserbase or Firecrawl?+
Tabstack positions its API around extracted data, cited research and completed public-web tasks. Its vendor comparison pages distinguish that workflow from managed browser infrastructure such as Browserbase and crawling or ingestion tools such as Firecrawl. Treat those comparisons as vendor positioning; choose by the required output, access boundary and control over the browser rather than assuming one service is universally better.
How does Tabstack pricing work?+
Tabstack uses credits per action. Its pricing documentation checked October 27, 2026 lists 10 credits for Markdown extraction, 50 for JSON extraction, 100 for Generate or Automate actions, and 250 for Fast or 350 for Balanced research actions. Research and automation can spend multiple actions per call; plan eligibility, overages and spending controls also affect the final cost.
Can Tabstack handle authenticated or dynamic pages?+
Hosted Tabstack Automate can interact with dynamic public websites, but it cannot log in or work behind an authenticated portal under the official guide checked October 27, 2026. Interactive input is for requested form values, not a promise of authenticated access. Pilo is the separate self-hosted option for automation in a browser you control; check its configuration and permissions for your intended workflow.