earakely-scale/JellyfinEnv-REA

jellyfin

jellyfin

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

Tables

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

Run it

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

Jellyfin on AgentEnv, from a REA capture

What this is. An environment of Jellyfin 10.10.7 (GPL-2.0), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (38 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, no write tool is offered because REA could confirm none, and data/reset restores it. Nothing executes Jellyfin. 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

• 38 tools (32 GET, 5 POST, 1 DELETE) over 96 recorded exchanges on http://127.0.0.1:3224; no recorded group excluded. 32 are offered; the other 6 are write tools the environment cannot apply (post_displaypreferences_usersettings, delete_items_by_hash, post_items_by_hash, post_playlists, post_playlists_items_by_hash, post_userfavoriteitems_by_hash), left out rather than faked.
• Path parameters: 18 hex. 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 38 tools.
• State: 41 entities of 12 types. 26 read tools answer from it and 6 replay their recorded responses. 6 list tools also answer filters, search, sorting or paging from it. No write tool is served.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/jellyfin-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.
• 24 retrieval 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_6f1c993cad742b377b2f43d9ca7c76a0d7af59b944d788ae3eaba8715da22b1f (scoped HAR sha256 9424597c22a8e05de3af93ea34521b543bea01b39744079ec1e881c327e058d5, 246,920 bytes), manifest view ev_375e42623b70e13faff660f8d092aeaf15ffc6be31a2727f5591d4825ab958b0, confidence inferred, authority historical-reference. The REA build is rea-agents 6.7.0 from main 3dbe314a with open pull requests merged: morluto/rea#5600 at 325c128f (which carries #5351 and #5352), #5787 at 9b868c07, #5800 at 35f3f2b5, #6136 at a477b21d, #6138 at 04b24a3e and #6169 at 733833ef. 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. No write tool is offered: its playlist create posts to /Playlists while every read of a playlist goes through /Items, so the playlists the capture made never became state and the create, the metadata update and the delete have no state target, so REA confirmed no write's effect and the environment is read-only. 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_users_items_by_hash	get_users_items_by_hash	hash=01ad108d20d3481e9e15fd3d6997c18f, SortBy=SortName, SortOrder=Ascending, IncludeItemTypes=MusicAlbum, Recursive=true, Fields=PrimaryImageAspectRatio,SortName, ImageTypeLimit=1, EnableImageTypes=Primary,Backdrop,Banner,Thumb, StartIndex=0, Limit=100, ParentId=911cb375d4dbd9806dd58ec6b6cee9e9	Items[0].UserData.Key	6300cdb3-69f4-c149-85c6-df46417de861
qa-02-get_users_items_by_hash_hash_2	get_users_items_by_hash_hash_2	hash=01ad108d20d3481e9e15fd3d6997c18f, hash_2=911cb375d4dbd9806dd58ec6b6cee9e9	Name	Lighthouse Rota Music
qa-03-get_playlists_items_by_hash	get_playlists_items_by_hash	hash=2774589b45efad94f4119ea6792461d2, Fields=PrimaryImageAspectRatio, EnableImageTypes=Primary,Backdrop,Banner,Thumb, UserId=01ad108d20d3481e9e15fd3d6997c18f	Items[0].Name	Tide Table Waltz
qa-04-get_users_items_latest_by_hash	get_users_items_latest_by_hash	hash=01ad108d20d3481e9e15fd3d6997c18f, Limit=16, Fields=PrimaryImageAspectRatio,Path, ImageTypeLimit=1, EnableImageTypes=Primary,Backdrop,Thumb, ParentId=37c97fe02330440cdf853417ba241c80	[0].Name	Quayside Market Ledger
qa-05-get_items_ancestors_by_hash	get_items_ancestors_by_hash	hash=692714480b0d4d767aa4fae2fa5ad545, userId=01ad108d20d3481e9e15fd3d6997c18f	[0].SortName	lighthouse rota suite
qa-06-get_items_by_hash	get_items_by_hash	hash=6300cdb369f4c14985c6df46417de861, userId=01ad108d20d3481e9e15fd3d6997c18f	Overview	Three tunes the Ada Finch Quartet plays for the winter rota.
qa-07-get_items	get_items	userId=01ad108d20d3481e9e15fd3d6997c18f, parentId=6300cdb369f4c14985c6df46417de861, sortBy=IndexNumber, fields=Overview,Genres	Items[0].Name	Granite Ledger
qa-08-get_library_virtualfolders	get_library_virtualfolders	no arguments	[0].Name	Copperfield Annex Reading Room
qa-09-get_items_latest	get_items_latest	userId=01ad108d20d3481e9e15fd3d6997c18f, parentId=911cb375d4dbd9806dd58ec6b6cee9e9, limit=5	[0].Name	Saltmarsh Depot Sessions
qa-10-get_musicgenres_kitchen_20jazz	get_musicgenres_kitchen_20jazz	userId=01ad108d20d3481e9e15fd3d6997c18f	Name	Kitchen Jazz
qa-11-get_artists_ada_20finch_20quartet	get_artists_ada_20finch_20quartet	userId=01ad108d20d3481e9e15fd3d6997c18f	UserData.Key	Artist-Ada Finch Quartet
qa-12-get_artists	get_artists	userId=01ad108d20d3481e9e15fd3d6997c18f, fields=Overview	Items[0].Overview	Lamp-room quartet led by Ada Finch, keeper of the tide tables.

Episodes

model	retrieval episodes	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	12	100%	1.0	19,811
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	23,245

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

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/JellyfinEnv-REA --bundle jellyfin --task smoke          # Docker, no model
agent-env hf run earakely-scale/JellyfinEnv-REA --bundle jellyfin --eval jellyfin-retrieval --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. Jellyfin ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (admin rea-admin from the startup wizard and two more users (Ada Finch, Rui Castellanos); libraries Lighthouse Rota Music, Saltmarsh Depot Photos and Copperfield Annex Reading Room over generated files; artists Ada Finch Quartet and Blue Heron Kitchen Band with overviews; albums Lighthouse Rota Suite (Coastal Folk, 3 tracks) and Saltmarsh Depot Sessions (Kitchen Jazz, 2 tracks) with overviews and tags; a photo album of 3 pictures with overviews; books The Rye Cracker Almanac and Quayside Market Ledger with overviews and tags; playlist Night Watch Mix (3 tracks); collection Granite Ledger...); people and projects 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 objects per write target (created, then updated and deleted where the app allows), reading each write back at once; the HAR was scoped to the API, the login exchange 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; 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	Jellyfin (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)	32 tools mirroring the application's endpoints
State	Full: a virtual clock, triggers, frozen windows, agents that answer	A state extracted from the capture (41 entities); read-only; 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; 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 Jellyfin; 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 Jellyfin instance over synthetic data; Jellyfin is GPL-2.0 software by the Jellyfin 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.