Skip to main content
View source

LlamaParse

View as Markdown

A RocketRide data node that sends documents to LlamaParse and writes flattened Markdown text and extracted tables into the pipeline; use it for readable content, not audit-grade source provenance.

About LlamaParse​

LlamaParse is a document-parsing service used to turn files into text and structured document content. This node uses its Python client to parse the document bytes supplied by a RocketRide pipeline and to return the resulting text and table data.

What it does​

The node processes incoming document data on tags and emits parsed Markdown on text plus tables on table. It isolates each LlamaParse request in a separate thread and event loop, which is useful where the surrounding runtime already has event-loop state. Choose it when flattened Markdown is the desired downstream format; use a structure-preserving parser when later review needs page boundaries, coordinates, or the original table-cell grid.

Lanes​

Lane inLane outDescription
tagstextParsed document text, flattened to Markdown.
tagstableTable content detected in the parsed result.

Configuration​

Start with an API key and the default simple configuration. Simple mode builds the LlamaParse constructor arguments from the visible controls; advanced mode instead parses the supplied JSON and merges its entries into those arguments.

Advanced Configuration​

Enable Advanced Configuration only when the simple controls cannot express the parsing options you need. Its value must be valid JSON and must not set api_key, which the node ignores when merging the advanced object. Save-time validation only warns about unknown keys or invalid JSON, but startup aborts when advanced mode has no configuration or when JSON cannot be parsed. A good value is an object containing only the constructor options you intend to override, such as {"parse_mode": "parse_page_with_llm"}.

Parse Mode and LVM Model​

The default Parse Mode, parse_page_with_lvm, passes that mode to the client, optionally passes LVM Model as vendor_multimodal_model_name, and sets a page-error tolerance of 0.05. The agentic and agentic_plus choices both select parse_page_with_agent and can use the same LVM model setting.

The displayed cost_effective (LLM) value does not match the code path that checks for cost_effective; as written, choosing it supplies no parse-mode argument. Keep the default or use an agentic choice when a specific mode must be sent. The LVM model is not used by the cost-effective path.

Additional Instructions and spreadsheet tables​

Additional Instructions are passed only for the legacy parse_page_with_lvm path, when non-blank; use them for parser guidance that belongs with that mode. Extract Sub Tables defaults to off and, when on, passes spreadsheet_extract_sub_tables=True to LlamaParse. Enable it only for spreadsheets whose nested tables need separate parsing; it has no special handling in the node beyond forwarding that setting.

Authentication​

Set API Key to the LlamaParse service key. Missing keys generate a save-time warning and cause startup to raise before a parser is created.

Requirements​

The service metadata declares the gpu capability. The node itself creates a LlamaParse client and sends document bytes to it; it does not download or run a local model. A working network connection to the service and its Python dependency are therefore required by this implementation.

Notes​

Scale-header preservation​

Financial statements declare their magnitude in a caption above the table ("(In millions)", "$ in thousands", "amounts in millions of USD"). If that caption is separated from its table by downstream chunking or an LLM step, every figure reads as raw units and is silently 1,000x, 1,000,000x, or 1,000,000,000x off. Nothing downstream catches it, because a wrong-by-1e6 number looks exactly like a right one.

To keep that loss visible, the parser post-processes the extracted Markdown:

  • Detects scale captions the parser already emitted ("(in millions)", "in thousands, except per share amounts", "(₹ in crore)", ...) and normalizes each to a unit, a numeric factor, and an optional currency. Prose uses of the same words ("millions of users", "billions served", "hundreds of thousands of dollars") are not treated as captions.
  • Welds a marker to the table. When a caption is in scope for a numeric table (same page/section, no page break between them), a normalized line is injected directly above the table, for example > Scale: amounts in millions (x1,000,000). A later chunker or LLM cannot separate the scale from its figures. A caption is only welded onto a table that actually looks numeric, so a caption sitting over a prose or roster table is never mis-attached.
  • Flags a missing scale, when the context is financial. When a numeric table has no caption in scope and the document shows a financial signal (a caption detected elsewhere, or a currency marking in the table itself), a warning line is injected above it: > [scale?] Scale not detected for this table. Figures may be in thousands, millions, or billions. Verify against the source. The warning errs toward caution: a visible warning is far cheaper than a silent 1e6 error. Plain numeric tables with no financial signal (inventory counts, server metrics, sports standings) are left untouched, so ordinary documents pay nothing for this feature. The markers are plain ASCII, so they survive a cp1252 console without an encoding error.

The welded markers travel in-band with the text and table lanes, so they are the durable downstream signal: whatever chunks or summarizes the output carries the scale (or the warning) with the figures.

For observability, the same results are also recorded in parsing_metadata (debug-logged by the instance; not currently re-emitted onto a lane):

  • detected_scales: list of {phrase, unit, factor, currency} for each caption found.
  • scale_warnings: per-table {table_index, status, ...} where status is scale_detected, scale_missing, not_financial, or already_annotated (the table already carried one of our markers and was left as-is).

Annotation is idempotent (re-running does not double-inject; the injected markers are not themselves read back as captions) and never blocks parsing: any failure falls back to the untouched text.

Out of scope. Recovering a caption that LlamaParse dropped from the source entirely is not handled here. That needs an independent source-text extraction and belongs to the audit-grade datalab_parse node. This node makes an emitted-but-separable scale un-loseable and flags its absence; it does not re-parse the source.

Flattened output and failures​

The node concatenates the returned document text and reads tables from either structured_data or items metadata. It does not retain page boundaries, bounding boxes, or a table HTML cell grid. Parsing errors, absent input, and import errors return empty text and structured data rather than an exception.

For files over 50 MiB the isolated call raises the parser timeout settings to at least 600 seconds. The worker waits up to 300 seconds for smaller files, 600 seconds above 100 MiB, and 900 seconds above 500 MiB; a timed-out worker returns an empty result with error metadata while its daemon thread is left to finish.

Upstream docs​

Schema​

FieldTypeDescriptionDefault
llamaparse.advanced_configstringAdvanced Configuration (JSON)
Enter configuration options in JSON format. For more information, see: LlamaParse Documentation		"{\n \"parse_mode\": \"parse_page_with_llm\",\n \"spreadsheet_extract_sub_tables\": false,\n \"system_prompt_append\": \"\",\n \"lvm_model\": \"anthropic-sonnet-4.0\"\n}"
llamaparse.api_keystringAPI Key
Your LlamaIndex API key for LlamaParse service
llamaparse.lvm_modelstringLVM Model
The LVM model to use for parsing when LVM or agentic modes are selected.		"anthropic-sonnet-4.0"
llamaparse.parse_modestringParse Mode
The parse mode to use for chosing complexity of the parse		"parse_page_with_lvm"
llamaparse.spreadsheet_extract_sub_tablesbooleanExtract Sub Tables
Extract sub-tables from spreadsheets for better table parsing.		false
llamaparse.system_prompt_appendstringAdditional Instructions
Additional instructions to append to the system prompt for LlamaParse.
llamaparse.use_advanced_configbooleanAdvanced Configuration
Check to use advanced JSON configuration instead of simple options.		false
llamaparse.use_system_prompt_appendbooleanUse Additional Instructions
Check to add custom instructions to the system prompt for LlamaParse.		false

Dependencies​

  • llama-parse
  • llama-index-core
  • llama-cloud
Last updated on