The project describes a multi-agent lawn-care advisory for Zeon Zoysia in coastal South Carolina that retrieves cited knowledge-base passages and provides recommendations about lawn-care timing. [1]
Verification: not independently verified These source claims are documented, not independently tested results.
The agricultural problem
The documented example asks whether it is too early to put down pre-emergent. [2]
What the AI agent does
The documented --ask command accepts the example pre-emergent timing question. [2]
Expected output
The README states that each chemical recommendation must cite an authoritative passage from the retrieved knowledge base that supports the claim; if no source is found, the system refuses the recommendation and tells the user where to look instead. [3]
Inputs and data
- The corpus ingester reads PDFs placed in
data/corpus/and public pages listed underseed_urls:inconfig.yaml. [4]
Workflow
- Run
uv run lawn-agents --ask "Is it too early to put down pre-emergent?"to pose the documented example question. [2] - For any chemical recommendation, the README requires a supporting authoritative passage from the retrieved knowledge base; if no source is found, the system refuses and tells the user where to look instead. [3]
Tools and setup
- The README setup instructions copy
.env.exampleto.envand specify configuring aGEMINI_API_KEY, or switching to Anthropic inconfig.yamland usingANTHROPIC_API_KEY. [5]
Costs
Unknown — the reviewed source does not verify a cost for this workflow.
Limitations
- The README says
--review-additionsis a stub, automatically ingested passages are flagged[unreviewed], and there is no promote/reject UI yet. [6]
Source evidence
Documentation-backed source excerpts, not independent tests of the product. Expand a source to inspect its exact recorded excerpt.
Source [1]DocumentedA multi-agent lawn-care advisory for Zeon Zoysia in coastal South Carolina; it retrieves cited passages from a local…
A multi-agent lawn-care advisory for Zeon Zoysia in coastal South Carolina. It pulls live weather and soil temperature, retrieves cited passages from a local knowledge base (Clemson HGIC, Super-Sod Lawn Academy, the Turfgrass Group, your own paid guides), and recommends when to fertilize, treat for grubs, apply pre-emergent, and put down micronutrients — with citations on every chemical recommendation.
Open original documentation ↗Source [2]DocumentedThe documented --ask example is uv run lawn-agents --ask "Is it too early to put down pre-emergent?".
6. Ask away uv run lawn-agents --ask "Is it too early to put down pre-emergent?" uv run lawn-agents --scheduled uv run lawn-agents --plan-year 2026 ```
Open original documentation ↗Source [3]DocumentedEvery chemical recommendation must cite an authoritative passage from the retrieved knowledge base that supports the claim;…
> Safety rule. This system never guesses on herbicide, fertilizer, or > insecticide products. Every chemical recommendation must cite an > authoritative passage from the retrieved knowledge base — and that > citation must actually support the claim, not merely accompany it > ([ADR 0010](docs/adr/0010-claim-support-validation.md)). If no source > is found, it refuses and tells you where to look instead.
Open original documentation ↗Source [4]DocumentedThe corpus ingester reads PDFs placed in data/corpus/ and public pages listed under seed_urls: in config.yaml.
1. *data/corpus/.pdf** — drop any PDFs you've purchased or downloaded here. Gitignored. 2. seed_urls: in config.yaml — public pages we recommend pre-loading (Super-Sod Lawn Academy, NCSU TurfFiles, etc.). Fetched, chunked, and embedded into the local LanceDB index. The URL list is yours to edit.
Open original documentation ↗Source [5]DocumentedThe setup instructions say to copy .env.example to .env and configure a GEMINI_API_KEY, or switch to Anthropic in…
3. Configure cp .env.example .env Edit .env and paste your GEMINI_API_KEY from aistudio.google.com (free-tier works). Or switch to Anthropic in config.yaml and use ANTHROPIC_API_KEY instead — see docs/adr/0006-provider-abstraction.md.
Open original documentation ↗Source [6]Documented--review-additions is a stub; automatically ingested passages are stored and flagged, and there is no promote/reject UI yet.
> Not yet implemented: --review-additions is currently a stub. Auto > ingested passages are stored and flagged, and the synthesizer labels > them [unreviewed] in its <sources> block so the human-in-the-loop > is preserved at the point of use — but there is no promote/reject UI > yet. Until there is, unreviewed passages accumulate in the index.
Open original documentation ↗