tabstack

# Tabstack — Web & PDF Tools

Tabstack is a web execution API for reading, extracting, transforming, and interacting with web pages and PDF documents. It handles JavaScript-rendered sites, structured data extraction, AI-powered content transformation, and multi-step browser automation.

## Available Tools

### 1. `tabstack_extract_markdown` — Read a page or PDF as clean Markdown

Best for: reading articles, documentation, PDF reports. This is the cheapest operation — prefer it when you just need to read content.

### 2. `tabstack_extract_json` — Pull structured data from a page or PDF

Best for: prices, product details, tables, invoices, any document with predictable repeating structure. Pass a JSON Schema defining the structure you want to extract.

### 3. `tabstack_generate` — Transform web/PDF content into a custom JSON shape

Best for: summaries, categorization, sentiment analysis, reformatting. Unlike `tabstack_extract_json` (which pulls existing data), `tabstack_generate` uses an LLM to *create* new content from a page. May be slower due to LLM processing.

### 4. `tabstack_automate` — Multi-step browser task in natural language

Best for: tasks needing real browser interaction — clicking, navigating across pages, filling forms. Also has built-in web search — when no URL is given, it searches the web itself. This makes it great for quick factual lookups (addresses, hours, prices) without needing a URL.

Takes 30-120 seconds.

**Interactive form-fill.** To let the browser agent fill a form that needs personal data, set `interactive: true` and pass the values in `data`, keyed by field label:

```json
{"task": "Sign up for the newsletter", "url": "https://example.com",
 "interactive": true,
 "data": {"Email": "ada@example.com", "Full name": "Ada Lovelace"}}
```

When the browser hits a form it answers from `data`. Each submission is shown to the user for **confirmation** before any personal data is sent — so don't put secrets you wouldn't want surfaced in a prompt into `data`. Behavior:

- **Match by field label** (case-insensitive). `data` keys should mirror the form's field labels.
- **Missing required field** → that form request is cancelled and the missing field name(s) are reported back in the result. Gather the values (ask the user) and re-run with them added to `data`.
- **User declines the confirmation** → the submission is cancelled and reported.

Gather the data you can up front (from the user, USER.md, or the vault) before calling — the missing-field report is a fallback, not the happy path. Without `interactive: true`, `tabstack_automate` never prompts and behaves as a normal read/search/click task.

### 5. `tabstack_research` — AI-powered deep web research

Searches the web, analyzes multiple sources, and synthesizes a comprehensive answer with citations. For simple factual lookups, `tabstack_automate` without a URL is faster and cheaper. Use `tabstack_research` when you need depth, multiple perspectives, or cited sources.

Supports `mode` parameter: `fast` for quick single-source answers, `balanced` (default) for deeper multi-source research.

Takes 60-120 seconds.

## Choosing the Right Tool

| Tool                        | Use when...                                     | Cost    | Speed   |
|-----------------------------|------------------------------------------------|---------|---------|
| `tabstack_extract_markdown` | Read/summarize a page or PDF                   | Lowest  | Fast    |
| `tabstack_extract_json`     | Structured data from a page or PDF             | Medium  | Fast    |
| `tabstack_generate`         | AI-transformed content from a page or PDF      | Medium  | Medium  |
| `tabstack_research`         | Answers from multiple web sources              | Medium  | 60-120s |
| `tabstack_automate`         | Browser interaction or simple web search       | Highest | 30-120s |

**Prefer cheaper operations when they suffice.** Use `tabstack_extract_markdown` for simple reading. Only use `tabstack_automate` when the task requires clicking, navigating, form interaction, or a quick web search without a known URL.

Inform the user before triggering multiple `tabstack_automate` calls — they are the most expensive.

## Error Handling

- On `tabstack_automate` failures, retry once. If it fails again, fall back to `tabstack_extract_markdown` for read-only tasks.
- If `tabstack_extract_json` returns poor results, try providing a more specific JSON Schema or switch to `tabstack_generate` with natural language instructions.
- If a page returns empty content, the site may block automated access — try `tabstack_automate` which uses a full browser.