AgroRAG answers herbicide, weed-management and spraying questions using your technical documents, citing the document and section behind each answer. [1]
Verification: README documented; not independently verified. These source claims are documented, not independently tested results.
The agricultural problem
Answers to herbicide, weed-management and spraying questions need to be traceable to the product labels, field guides or other technical documents being used. [1]
What the AI agent does
AgroRAG embeds each question, retrieves the closest document chunks from pgvector and discards chunks below a similarity threshold. [3]
Expected output
The web UI offers clickable citations that highlight the source passage supporting an answer. [4]
Inputs and data
- Place PDFs,
.mdor.txtfiles indata/docs/and runreindexar.bat, or add them through the web UI’s + Documento control. [2]
Workflow
- Ask a question after the documents are indexed; AgroRAG compares it with stored chunks and filters out weak matches. [3]
- Illustrative example: Ask what your field guide says about a weed-management question, then check the cited document and section before relying on the answer. [1]
- Without an API key, inspect the cited extractive passages rather than a generated answer; the same mode is used if the LLM fails. [4]
Tools and setup
- Double-click
instalar.batto create the environment, install dependencies and index sample documents; then launch the app withiniciar.bat. [5]
Costs
Unknown — the reviewed source does not verify a cost for this workflow.
Limitations
- Answers depend on the loaded documents and do not replace the product label or advice from an agronomist. [6]
Practical details from the source
These additional notes summarize the linked README for this exact documented revision. They do not represent a field trial or change the verified draft.
Local or database-backed: The README offers a local NumPy index without setting up PostgreSQL, or a pgvector-backed database for storage. [4]
An example to try: Load a weed-management guide, ask what its sections say about resistance management, and compare the answer with the cited original passage before making any field decision. [1] The tool does not replace the product label or a qualified agronomist. [6]
API key and cost: Extractive mode works without a Groq API key. [4] Costs for optional hosted services are not verified by this guide.
macOS / Linux installation (from the project README)
These commands come directly from the README’s macOS / Linux Quick start ↗, not from the Windows batch-file instructions.
python -m venv .venv && source .venv/bin/activate pip install -r requirements.txt cp .env.example .env python -m app.cli ingest uvicorn app.main:app --reload
Adding a Groq key to the local environment file is optional. Consult the linked README for configuration details.
Source evidence
Documentation-backed source excerpts, not independent tests of the product. Expand a source to inspect its exact recorded excerpt.
Source [1]DocumentedAgroRAG answers questions about herbicides, weed management and spraying, using your own technical documents (datasheets,…
AgroRAG answers questions about herbicides, weed management and spraying, using your own technical documents (datasheets, product labels, field guides). Every answer cites the exact document and section it came from, so it can be checked instead of trusted blindly.
Open original documentation ↗Source [2]DocumentedDrop PDFs, .md or .txt files into data/docs/ and run reindexar.bat, or use + Documento in the web UI.
Drop PDFs, .md or .txt files into data/docs/ and run reindexar.bat, or use + Documento in the web UI.
Open original documentation ↗Source [3]Documented1. Ingestion: PDFs and Markdown files are parsed, split into section-aware chunks (headings are kept as metadata, overlap…
1. Ingestion: PDFs and Markdown files are parsed, split into section-aware chunks (headings are kept as metadata, overlap never starts mid-sentence) and embedded. 2. Retrieval: the question is embedded and the closest chunks are fetched from pgvector (cosine distance, HNSW index). Chunks below a similarity threshold are discarded, so off-topic questions get an honest "I don't have that information" instead of a hallucination.
Open original documentation ↗Source [4]Documented• Web UI with clickable citations that highlight the source passage - Upload new PDFs from the UI; the index is rebuilt…
• Web UI with clickable citations that highlight the source passage • Upload new PDFs from the UI; the index is rebuilt automatically • Two vector stores behind one interface: pgvector (production / Neon) or a local NumPy index (zero setup) • Works without an API key: an extractive mode returns the most relevant passages, already cited • Graceful degradation: if the LLM fails (bad key, rate limit) it falls back to extractive mode; if the embedding model can't be downloaded it falls back to an offline hashing embedder
Open original documentation ↗Source [5]Documented1. Double-click instalar.bat: creates the virtualenv, installs dependencies and indexes the sample documents. 2. (Optional)…
1. Double-click instalar.bat: creates the virtualenv, installs dependencies and indexes the sample documents. 2. (Optional) Open .env and set GROQ_API_KEY (free at console.groq.com). 3. Double-click iniciar.bat: the app opens at http://localhost:8000.
Open original documentation ↗Source [6]DocumentedAgroRAG is a decision-support tool. Answers are only as good as the loaded documents and do not replace the product label or…
AgroRAG is a decision-support tool. Answers are only as good as the loaded documents and do not replace the product label or advice from an agronomist.
Open original documentation ↗