earakely-scale/UmamiEnv-REA

umami

umami

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

Tables

• umami_tasks: one row per task of the bundle, with its prompt, envs, verifiers and steps.
• umami_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/umami.jsonl holds each run's record and native trajectory, and bundles/umami/ the bundle itself.

Run it

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

Umami on AgentEnv, from a REA capture

What this is. A replay environment of Umami 3.4 (MIT), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (49 tools), and rea environment-export wrote the bundle you can run from this dataset. Every tool call replays a retained record from the capture, cites the record and the Evidence it came from, and nothing executes Umami. Tasks are graded deterministically, with no judge.

What's in it

• 49 tools (46 GET, 2 POST, 1 DELETE) over 100 recorded exchanges on http://127.0.0.1:3215; no recorded group excluded.
• Path parameters: 44 uuid. REA's four format rules (numeric, uuid, hex, iso-date) applied unconditionally; the three opt-in rules (after-uid-literal, opaque-token, sibling-variance) were requested and are labelled per parameter. With the four format rules alone the manifest has 49 tools.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/umami-retrieval.toml): each asks one factual question about a seeded object, with a distinct answer that the app does not ship by default, answerable through a recorded exchange; reward 1 if the response contains the recorded value, else 0. Plus smoke, which deploys the environment and checks replay through the MCP endpoint.
• 24 recorded episodes 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_12a4e7f97b75d77076ad5f7c072a7073a32f0c713f8488096e727a134cc78d35 (scoped HAR sha256 f896fb119afe85ef513002099f9c2762402ec54cde9878bb58ee374824d67fc4, 151,479 bytes), manifest view ev_6559a1b1ad3429225eecffeee0b3397adeb0a5cb9535d4de80544f3c9e175ead, confidence inferred, authority historical-reference. The REA build is a prerelease of the environment commands proposed upstream in morluto/rea#4954, #4955, #4975 and #4999; the bundle installs it from the tarball it ships, so the environment runs without waiting for an npm release.

What replay means, and does not

A tool call returns the recorded response whose path and query values best match the arguments, with replay.exact_match telling you whether it matched exactly or fell back to the closest record. Reads of seeded objects behave like the application. Writes return the recorded answer of the one write the session made, and the environment has no state: a created object does not appear in a later list. This is what makes retrieval tasks hard-gradable and write tasks weak; the manifest's unknowns say so.

Tasks and grading

task	tool	arguments	field	answer
qa-01-get_teams_websites_by_uuid	get_teams_websites_by_uuid	uuid=14230b65-bafc-4474-9548-d6b5d0b8f066	data[0].name	Lighthouse Rota Portal
qa-02-get_websites_segments_by_uuid	get_websites_segments_by_uuid	uuid=3c2d6c7b-7911-42ee-9826-3290f29fd2f6, type=segment	data[0].name	Kelp Broth Readers
qa-03-get_teams_by_uuid	get_teams_by_uuid	uuid=14230b65-bafc-4474-9548-d6b5d0b8f066	name	Saltmarsh Depot Analytics
qa-04-get_websites_session_data_values_by_uuid	get_websites_session_data_values_by_uuid	uuid=8a5e3955-ffbd-4271-80d9-50cb2644b7cd, startAt=1791754915801, endAt=1792359715801, propertyName=crew	[0].value	day shift
qa-05-get_websites_by_uuid	get_websites_by_uuid	uuid=629e163c-338b-4fc7-ab93-dccad214ae8d	name	Granite Ledger Docs
qa-06-get_websites_funnels_by_uuid_uuid_2	get_websites_funnels_by_uuid_uuid_2	uuid=8a5e3955-ffbd-4271-80d9-50cb2644b7cd, uuid_2=5d3809f4-d55e-4daa-904b-d991cb9bdf70	description	From the rota page to a completed night-shift signup.
qa-07-get_websites_sessions_by_uuid	get_websites_sessions_by_uuid	uuid=3c2d6c7b-7911-42ee-9826-3290f29fd2f6, startAt=1792270800000, endAt=1792360799999, unit=hour, timezone=UTC, maxResults=10000	data[0].os	Windows 10
qa-08-get_websites_funnels_by_uuid	get_websites_funnels_by_uuid	uuid=8a5e3955-ffbd-4271-80d9-50cb2644b7cd	data[0].name	Rota Signup Funnel
qa-09-get_websites_annotations_by_uuid_uuid_2	get_websites_annotations_by_uuid_uuid_2	uuid=629e163c-338b-4fc7-ab93-dccad214ae8d, uuid_2=19ba1a3e-5981-4d52-872c-1c6fd5c11a2a	note	Quarterly stone count published; audited by Rui Castellanos.
qa-10-get_websites_event_data_properties_by_uu	get_websites_event_data_properties_by_uuid	uuid=8a5e3955-ffbd-4271-80d9-50cb2644b7cd, startAt=1791754915801, endAt=1792359715801	[0].eventName	rota-signup
qa-11-get_pixels_by_uuid	get_pixels_by_uuid	uuid=386b57c4-c620-44e2-aff3-445d3df0746a	name	Granite Ledger Newsletter Pixel
qa-12-get_websites_goals_by_uuid_uuid_2	get_websites_goals_by_uuid_uuid_2	uuid=629e163c-338b-4fc7-ab93-dccad214ae8d, uuid_2=59d11a64-f552-482e-a850-21482144c18e	name	Ledger PDF Downloads

Episodes

model	tasks	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	11	92%	1.0	16,065
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	20,987

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

openai/gpt-oss-120b:groq missed qa-06-get_websites_funnels_by_uuid_uuid_2. Miss by gpt-oss-120b:groq on qa-06-get_websites_funnels_by_uuid_uuid_2: the model's answer did not contain the expected value.

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/UmamiEnv-REA --bundle umami --task smoke          # Docker, no model
agent-env hf run earakely-scale/UmamiEnv-REA --bundle umami --eval umami-retrieval --model openai/gpt-oss-120b:groq

The retrieval eval needs 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. Umami ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (2 teams (Saltmarsh Depot Analytics, Copperfield Annex Web), 3 websites (Lighthouse Rota Portal, Granite Ledger Docs, Blue Heron Kitchen), 2 users with team roles, 1 short link, 1 pixel, 1 board, 2 annotations, 1 segment, 1 goal, 1 funnel, and 86 synthetic collector calls (68 pageviews, 12 custom events with data, 6 identify calls) over 24 visitors in the last six days); people and projects are invented.
2. Playwright drove a login, the main pages and a scripted read, write, read sequence through the app's JSON API; the HAR was scoped to the API, the login exchange dropped, secret-looking body 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; gen_tasks.py derived the retrieval tasks from the manifest and the recorded bodies; 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	Umami (this dataset)
Environment	A simulation: a container quay, ships, cranes, weather, with A2A ship agents	A replay of a real application's API, captured from one session
Observation	Tool results computed by the simulator as the week unfolds	Recorded responses, cited to Evidence
Action space	Planning tools (check_plan, confirm_berths, advance, message_agent)	49 tools mirroring the application's endpoints
State	Full: a virtual clock, triggers, frozen windows, agents that answer	None: every call is independent; writes replay one recorded answer
Reward	Deterministic, against a CP-SAT optimum or a re-planner anchor; continuous 0 to 1	Deterministic, recorded-value match; 0 or 1 per task
What it measures	Planning under uncertainty and negotiation	Tool discovery, parameter inference and faithful retrieval in a real API
Data provenance	Real 2024 port calls and weather; simulated workloads	Synthetic seed data on a self-hosted Umami; 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 list exactly what a stateful version would still have to specify.

Licence and credits

Bundle code and tasks: Apache-2.0. Recorded content: responses generated by a self-hosted Umami instance over synthetic data; Umami is MIT software by Umami Software, Inc. 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.