Output formats | Parallel download and rate limits | The files describing a download | Updating an existing folder | Checking without downloading | All groups of a user | Helper functions | Library version | Last modification time | User id from user name | Groups of a user

Download a group | Update it

Why SPECTER2 for academic papers | One-time setup | Python environment | Run the merge script | Serving via TEI | Verify | Using from R | Smoke test | Where this fits | Notes

Connection and Authentication | Query Construction | Request Execution | Storage and Refresh | Content and Abstract Pipeline | Utility

Folder Contracts | Request/Parquet Workflow | Content/Abstract Workflow | Data Contracts | Base Parquet Corpus | Abstract Parquet | read_corpus(..., abstracts = TRUE) | Execution Contracts | Error-Handling Contracts | Stability Notes

End-to-End Corpus Workflow (Multiple Queries) | 1) Create connection and multi-query search input | 2) Fetch search results into endpoint-structured project folders | 3) (Optional) Explicit parquet conversion step | 4) Download linked source content | 5) Convert downloaded content to markdown | 6) Build query-level abstract parquet files | 7) Read corpus and link abstracts | Practical Notes

Enrich Endpoint: Context Discovery for Web and News | Create the connection once | Build web and news enrich queries | Execute a focused run for each enrich type | Move to a thematic batch workload | Handle request failures without losing the whole run | Convert enrich output to parquet | Operational recommendations

FastGPT Endpoint: Practical Prompt Workflows | Create one connection for all runs | Build a single prompt query | Execute and store the first response | Scale prompts to a batch job | Keep long runs resilient with graceful error handling | Convert FastGPT output to parquet | Operational recommendations

kagiPro: First End-to-End Workflow | Project-folder first workflow | Install and load the package | Create a secure API connection | Build your first search query | Execute and persist search results | Run one example from each endpoint | Enrich web | Enrich news | Summarize (text input) | FastGPT | Convert JSON results to parquet | Bridge to OpenAlex-style vector input | Where to go next | Session info

Search Endpoint: From Question to Reusable Search Pipeline | Start with a reusable connection | Build one precise search query | Execute the request and persist results | Scale from single query to query grid | Choose your failure policy explicitly | Convert search JSON to parquet for analysis | Operational recommendations

Summarize Endpoint: Robust Summaries in Real Pipelines | Establish a reusable connection | Build summarize queries for URL and text sources | Execute a normal summarize request | Handle known short-input failures safely | Run mixed batches with one success and one failure | Convert summarize JSON to parquet | Operational recommendations

Purpose | File Structure | High-Level Backend Flow | End-to-End Call Order in embed_corpus() | Data Contract | Input dataset expected by embed_corpus() | Canonical text construction | Output columns per embedded row | Shared Backend Rules | Authentication Model | Provider Responsibilities | Hugging Face (provider = "hf") | OpenAI (provider = "openai") | TEI (provider = "tei") | Function Examples | 1) HF backend (default-style) | 2) OpenAI backend | 3) Local TEI backend | How to Add a New Backend | Step 1: add dispatch entry in core | Step 2: create provider file | Step 3: enforce adapter contract | Step 4: docs and validation | Operational Flowchart for embed_corpus() | Troubleshooting | 401 / 403 from provider API | Row mismatch errors | Model metadata unavailable (dim = NA)

Why this workflow exists | OpenAI vs OpenAlex (important) | Lifecycle | Prerequisites | Step 1: Submit jobs | What preflight checks do | Where state is written | Step 2: Check status | Interpreting status values | Step 3: Collect completed jobs | Output compatibility guarantee | Common pitfalls and what to check | 1) Authentication failures (401) | 2) Single oversized records | 3) Duplicate submission risk | 4) Concurrent collectors | 5) Confusing labels | Data retention and expiry (OpenAI) | Recommended operating pattern | Monitoring checklist | References

Purpose | Package Philosophy | Function Groups | 1) Embedding backend abstraction | 2) Corpus embedding orchestration | 3) Text preparation | 4) Similarity and distance | 5) Supervised scoring and calibration | 6) Embedding-space visualization | Canonical Workflow | Data and Storage Model | Corpus input | Embeddings output | Dry-run output | Text-Preparation Contract | Why this contract matters | Decision Model: Geometric + Supervised | Minimal End-to-End Example | Operational Guidance | Related Vignettes | Summary

Introduction | Requirements | Notes on execution | TEI endpoint | Embed a small corpus | Compute prototype distances | Compute ridge distances and scores | Troubleshooting | Reproducibility

Purpose | Start TEI | Local binary | Alternative port | Verify service | Health | Info | Embed smoke test | Use TEI endpoint in package | Process management (shell) | Recommended production pattern | Troubleshooting | Port already in use | Empty/invalid responses | Slow throughput

Purpose | Where Cleaning Happens | Cleaner Contract | Cleaning Pipeline (Step by Step) | Cleaning Flow Diagram | Why These Rules | Mode Behavior (lenient, balanced, strict) | Example: Default Cleaning | Example: Before/After on Individual Cases | Example: Change Missing-Abstract Policy | Example: Use a Stricter Cleaning Configuration | Using Custom Patterns | Integration with embed_corpus() | Preprocessor Validation in embed_corpus() | Output Provenance and Auditability | Recommended Operational Defaults | Summary

Overview | Installation of plantuml binary | Plotting Plantuml graphics | Define plantuml code | Plotplantuml Object | Usage of the local server

Register the plantuml knitr engine | PlantUML Code Chunk Options | A simple chart as png | A simple chart as png and with code | A simple chart using format auto with code | Now only the code | And now as ASCII Art