earakely-scale/PhotoPrismEnv-REA

photoprism

photoprism

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

Tables

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

Run it

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

PhotoPrism on AgentEnv, from a REA capture

What this is. A replay environment of PhotoPrism 261007 (AGPL-3.0), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (13 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 PhotoPrism. Tasks are graded deterministically, with no judge.

What's in it

• 13 tools (10 GET, 1 POST, 1 DELETE, 1 PUT) over 277 recorded exchanges on http://127.0.0.1:3401; 21 recorded groups excluded (21 as UI pages or assets).
• Path parameters: 5 opaque-token, 1 hex. 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 28 tools.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 10 retrieval tasks (evals/photoprism-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.
• 20 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_28d39ee2f958aaba2513fdbac537632e40e82a4855acc760da1d2448023e0c60 (a capture_browser_scenario session REA recorded itself; capture file sha256 4210fe0f993bd5230688332e895dd860071b89c061f35eea92cef9bada99e31d, 3,155,176 bytes), manifest view ev_81f20a4edbd785a66b689f99eae57d55989b94f43432ad290e28fb9131bda1e0, confidence inferred, authority controlled-replay. The REA build is a prerelease of the environment commands proposed upstream in morluto/rea#4954, #4955, #4975, #4999 and #5182; 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_photos_by_uid	get_photos_by_uid	uid=ptn547rxew9tkh8d	Caption	Sky fading from pale blue to navy over the Saltmarsh Depot loading bay
qa-02-get_albums_by_uid	get_albums_by_uid	uid=atn547u7js7hz0g1	Description	Roof camera frames and tide boards from the Saltmarsh Depot spring survey.
qa-03-get_albums	get_albums	count=180, offset=0, order=favorites, reverse=false, type=album	[0].Slug	granite-ledger-colour-study
qa-04-get_photos	get_photos	count=156, offset=0, merged=true, camera=0, lens=0, year=0, month=0, order=newest, reverse=false, public=true, quality=3	[0].Caption	Gold lantern disc against a navy night above the Blue Heron Kitchen pier
qa-05-get_labels	get_labels	count=180, offset=0, order=relevance	[0].Slug	signal-paint
qa-06-get_photos_by_uid	get_photos_by_uid	uid=ptn547rxew9tkh8d	Files[0].Name	saltmarsh-depot/20240312_071500.png
qa-07-get_albums_by_uid	get_albums_by_uid	uid=atn547u7js7hz0g1	Slug	saltmarsh-depot-survey
qa-08-get_albums	get_albums	count=180, offset=0, order=favorites, reverse=false, type=album	[0].Title	Granite Ledger Colour Study
qa-09-get_photos	get_photos	count=156, offset=0, merged=true, camera=0, lens=0, year=0, month=0, order=newest, reverse=false, public=true, quality=3	[0].Title	Lantern moon over the heron pier
qa-10-get_labels	get_labels	count=180, offset=0, order=relevance	[0].Name	Signal Paint

Episodes

model	tasks	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	10	10	100%	1.0	7,481
zai-org/GLM-5.3-Flash:zai-org	10	10	100%	1.0	9,034

Every episode is in the episodes config with its transcript; rewards 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/PhotoPrismEnv-REA --bundle photoprism --task smoke          # Docker, no model
agent-env hf run earakely-scale/PhotoPrismEnv-REA --bundle photoprism --eval photoprism-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. PhotoPrism ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (site title: Saltmarsh Depot Archive; photos: 12 generated PNGs (gradients, stripes, rings, checks, discs; 200x160 to 640x427, 0.7-6 KB) in three originals folders (saltmarsh-depot, copperfield-annex, blue-heron-kitchen), each with a manual title, caption, keywords, notes, artist, copyright, licence, taken date (March-May 2024, UTC) and coordinates; 4 favourites; albums: Saltmarsh Depot Survey (favourite, 4 photos), Copperfield Annex Workshop (4), Blue Heron Kitchen Tasting Night (4), Granite Ledger Colour Study (favourite, 5); labels: Tide Gauge, Signal Paint, Water Pattern, Crate Lot,...); people and projects are invented.
2. REA's capture-browser-scenario launched Chromium and drove the app's own UI, with any credential given as a declared secret; it recorded every network transaction with credential headers and declared secrets redacted, and the environment-manifest view read that capture directly, with no HAR or scrubbing step.
3. 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	PhotoPrism (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)	13 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 PhotoPrism; 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 PhotoPrism instance over synthetic data; PhotoPrism is AGPL-3.0 software by PhotoPrism UG and 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.