earakely-scale/GrocyEnv-REA

grocy

grocy

An AgentEnv bundle with its recorded runs, published with agentenv-hf.

Tables

• grocy_tasks: one row per task of the bundle, with its prompt, envs, verifiers and steps.
• grocy_episodes: one row per recorded run, with its status, model, reward and scores, the prompt and response, and the agent's transcript as chat messages when its trajectory format is one agentenv-hf reads.

raw/grocy.jsonl holds each run's record and native trajectory, and bundles/grocy/ the bundle itself.

Run it

pip install agentenv-framework huggingface_hub
hf download earakely-scale/GrocyEnv-REA --repo-type dataset --local-dir agentenv-dataset
agent-env run agentenv-dataset/bundles/grocy

Grocy on AgentEnv, from a REA capture

What this is. An environment of Grocy 4.7.1 (MIT), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (43 tools) and an environment state, the entities the capture's reads showed, and rea environment-export wrote the bundle you can run from this dataset. Reads the state can render are answered from it, write tools whose effect the capture confirmed change it, and data/reset restores it. Nothing executes Grocy. Tasks are graded deterministically, with no judge.

v0.2.0 rebuilds this environment from a new capture of an instance seeded the same way, recorded so each write's effect is shown: two scratch objects per write target, every write read back at once. A REA build that serves a state, confirmed writes and list queries projects it; the tag v0.1.0 keeps the earlier replay-only version.

What's in it

• 43 tools (40 GET, 1 POST, 1 DELETE, 1 PUT) over 87 recorded exchanges on http://127.0.0.1:3222; no recorded group excluded. 43 are offered.
• Path parameters: 22 numeric. REA's four format rules (numeric, uuid, hex, iso-date) applied unconditionally; the four opt-in rules (after-uid-literal, opaque-token, sibling-variance, prefix-variance) were requested and are labelled per parameter. With the four format rules alone the manifest has 43 tools.
• State: 50 entities of 24 types. 39 read tools answer from it and 1 replays its recorded response. 4 list tools also answer filters, search, sorting or paging from it. Served write tools: post_objects_products (create), delete_objects_products_by_id (delete), put_objects_products_by_id (update).
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/grocy-retrieval.toml): each asks one factual question about a seeded object, with a distinct answer that the app does not ship by default; reward 1 if the response contains the recorded value, else 0. Plus smoke, which deploys the environment and checks it through the MCP endpoint.
• 3 write tasks (evals/grocy-writes.toml): each asks for a change in plain language and is graded from the environment's state through data/get, not from the agent's reply; reward 1 only if every check passes. Their grade is in scores.state; reward holds the retrieval verifier's score, so it is empty for them, as for smoke.
• 24 retrieval episodes and 6 write episodes, plus the smoke run, by openai/gpt-oss-120b:groq, zai-org/GLM-5.3-Flash:zai-org through the Hugging Face router, with their chat transcripts.
• Provenance: capture ev_48aee427d8032874d6387d747a2315de8322b856f2d564fa845765ea294edeb5 (scoped HAR sha256 5f76c929e0affda92371051f4bebf4b6f3fcef070de5fe5618deb50fdf230787, 203,064 bytes), manifest view ev_f3a18778086f84ac2fd786200f23c5e0cbe59c049b7463c01d3f8597c036f8e8, confidence inferred, authority historical-reference. The REA build is rea-agents 6.7.0 from main 7091be29 with open pull requests merged: morluto/rea#6385 at 28c3ae97. The bundle installs it from the tarball it ships, so the environment runs without waiting for an npm release.

What the state means, and does not

Reads the state can render, item, list and document views of the entities the capture's responses showed, answer from it. List reads filtered, searched, sorted or paged answer from it too wherever the capture's recorded calls explain what each query parameter does, including values no call recorded; a parameter the recordings do not explain, or a flag's unrecorded value, replays. Other reads replay the recorded response whose path and query values best match the arguments, with replay.exact_match saying whether it matched exactly; a replayed read does not change after a write. A served write was confirmed by the capture itself: replaying the session against the extracted state, a later recorded read reproduced its effect and none contradicted it. A create generates ids and clock values deterministically, so the same calls after a reset give the same state; an update changes the members the call sends; a delete removes the entity. Each takes only the body members the recorded call showed and refuses others with a message that names the accepted ones. Its effect shows in the reads the state renders, lists included wherever a list lists its whole collection or a learned query answers the call; a list call nothing explains keeps its recorded members, dropping a deleted entity but not adding a new one. The data plane at /agentenv takes data/reset (restore the extracted state), data/add and data/get (return it), which is how write tasks are graded.

Not modelled: side effects a write has beyond its own entity (counters, copies in other collections, related entities), preconditions the application enforces before a write, list query parameters the recorded calls do not explain, paging members such as page counts and next links when a call asks for another page size or page, and anything the capture never showed. The manifest's unknowns list what a capture cannot establish.

Tasks and grading

Retrieval

task	tool	arguments	field	answer
qa-01-get_stock_products_entries_by_id	get_stock_products_entries_by_id	id=2	[0].note	Batch SD-118 from the east pans
qa-02-get_objects_shopping_lists_by_id	get_objects_shopping_lists_by_id	id=2	name	Lighthouse Rota Night Shift
qa-03-get_stock_products_locations_by_id	get_stock_products_locations_by_id	id=3	[0].location_name	Copperfield Annex Larder
qa-04-get_objects_task_categories_by_id	get_objects_task_categories_by_id	id=2	name	Lamp Watch
qa-05-get_objects_recipes_by_id	get_objects_recipes_by_id	id=1	description	Kelp broth stew for the night shift, ready in forty minutes.
qa-06-get_objects_equipment_by_id	get_objects_equipment_by_id	id=1	description	Four-tray kelp dehydrator; the manual is kept in the annex office.
qa-07-get_objects_tasks_by_id	get_objects_tasks_by_id	id=1	name	Recount the Granite Ledger crates
qa-08-get_objects_shopping_locations_by_id	get_objects_shopping_locations_by_id	id=1	name	Blue Heron Kitchen Market
qa-09-get_objects_quantity_units_by_id	get_objects_quantity_units_by_id	id=4	description	Wooden crate of twelve, stamped with the depot mark.
qa-10-get_recipes_fulfillment_by_id	get_recipes_fulfillment_by_id	id=1	product_names_comma_separated	Smoked Kelp Broth,Granite Salt Flakes
qa-11-get_stock_products_by_id	get_stock_products_by_id	id=5	location.name	Granite Ledger Lockers
qa-12-get_objects_chores_by_id	get_objects_chores_by_id	id=2	name	Rotate the Kelp Barrels

Writes

Graded from the environment's state through data/get (scores.state in the episodes table; reward 1 only if every check passes).

task	prompt	checks
write-create	In this Grocy household, add a new product named "Saltmarsh Sea Biscuits" with the description "Hard biscuits for the night watch.", stored in the Granite Ledger Lockers, with Tin as its quantity unit for purchase, stock, consumption and price.	a product named Saltmarsh Sea Biscuits exists; it is stored in the Granite Ledger Lockers (location 4); Tin (unit 5) is its purchase, stock, consumption and price unit; the five seeded products are still there
write-update	In this Grocy household, change the description of the product "Heron Rye Crackers" to "Twice-baked rye crackers packed by Tomas Vey, sold in tins of twelve."	the Heron Rye Crackers have the new description; no product was added or removed; another product's description is unchanged
write-delete-lookup	In this Grocy household, one product holds the paraffin lamp oil kept in crates. Find it and delete it.	the Copperfield Lamp Oil product is gone; the other four products are still there; the other lamp supply, the Lamp Wick Bundle, is still there

Episodes

model	retrieval episodes	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	12	100%	1.0	17,971
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	21,084

Every episode is in the episodes config with its transcript; scores are the deterministic verifier scores.

model	write episodes	passed
openai/gpt-oss-120b:groq	3	3
zai-org/GLM-5.3-Flash:zai-org	3	3

Every recorded episode passed.

Run it yourself

uv tool install agentenv-framework --with 'agentenv-hf @ git+https://github.com/earakely-scale/agentenv-hf-plugin@v0.3.0'
agent-env hf run earakely-scale/GrocyEnv-REA --bundle grocy --task smoke          # Docker, no model
agent-env hf run earakely-scale/GrocyEnv-REA --bundle grocy --eval grocy-retrieval --model openai/gpt-oss-120b:groq
agent-env hf run earakely-scale/GrocyEnv-REA --bundle grocy --eval grocy-writes --model openai/gpt-oss-120b:groq

The evals need a model endpoint: set [model] base_url and api_key in .agentenv/config.toml (the Hugging Face router works with any hf_ token). The bundle's agents/replay-qa is a small OpenAI-compatible tool-calling agent; any A2A agent can replace it.

How it was made

1. Grocy ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (users Ada Finch and Rui Castellanos; 3 locations, 4 quantity units, 2 product groups, 2 stores; 5 products, each with one purchase; 2 shopping lists with 4 items; 3 chores, 2 batteries, 2 task categories with 2 tasks, 2 pieces of equipment and 1 recipe); people and places are invented.
2. Playwright drove a login, the main pages and a scripted read sequence through the app's JSON API, then wrote two scratch products (created, edited and deleted) and edited and deleted a seeded product, reading each write back at once by item and by the product list; the HAR was scoped to the API, the login exchange and the web app's own table-settings calls dropped, secret-looking string fields scrubbed, headers and cookies removed.
3. rea inspect-web-network-capture retained the exchanges as Evidence; rea inspect-analysis-view projected the environment-manifest view; rea environment-export wrote this bundle and the state it serves; gen_tasks.py derived the retrieval tasks from the manifest and the recorded bodies; the write tasks were written by hand against the served write tools and the seeded names; agent-env run recorded the episodes; agent-env hf publish wrote this dataset. Scripts: https://github.com/earakely-scale/rea-environments.

Compared with PortSimEnv on AgentEnv

	PortSimEnv-AgentEnv	Grocy (this dataset)
Environment	A simulation: a container quay, ships, cranes, weather, with A2A ship agents	An application's API rebuilt from one captured session
Observation	Tool results computed by the simulator as the week unfolds	Answers rendered from the extracted state, or recorded responses, cited to Evidence
Action space	Planning tools (check_plan, confirm_berths, advance, message_agent)	43 tools mirroring the application's endpoints
State	Full: a virtual clock, triggers, frozen windows, agents that answer	A state extracted from the capture (50 entities); served writes change it; data/reset restores it
Reward	Deterministic, against a CP-SAT optimum or a re-planner anchor; continuous 0 to 1	Deterministic: recorded-value match for retrieval, state checks for writes; 0 or 1 per task
What it measures	Planning under uncertainty and negotiation	Tool discovery, parameter inference and faithful retrieval in a real API, and changing its state
Data provenance	Real 2024 port calls and weather; simulated workloads	Synthetic seed data on a self-hosted Grocy; nothing real
How it was built	Hand-written plugin, months of work	Generated from a capture in one scripted pass
Where it is strong	Long-horizon decisions with hard rewards	Breadth: any app you can capture becomes an environment

The two are complementary: PortSimEnv shows what a hand-built simulation can grade; this dataset shows how far a capture alone gets you, and the manifest's unknowns and the state's limits above say what it still leaves out.

Licence and credits

Bundle code and tasks: Apache-2.0. Recorded content: responses generated by a self-hosted Grocy instance over synthetic data; Grocy is MIT software by Bernd Bestel and the Grocy contributors and is not redistributed here beyond the recorded JSON responses. REA is MIT software by morluto; AgentEnv Framework is Apache-2.0 by Scale AI.