REA Environments on AgentEnv
50 reinforcement-learning environments, each built by REA from one recorded browser session and packaged as an AgentEnv bundle. An agent gets the application's API as MCP tools; every call replays a response the real application gave during the session, with the record and Evidence it came from. Nothing executes the application, so episodes are fast, cheap and deterministic.
How each one is made
What is in the collection
Baseline results
One run per task per model with a small tool-calling agent; reward 1 when the answer contains the recorded value. Each dataset carries every episode with its transcript, and its card explains each miss.
Browse the environments
Gitea 28.1.0
Git hosting, issues and code review
77 tools · 97 exchanges · 12 tasks
MIT
Grafana 13.2.4
observability dashboards, alerting and data sources
48 tools · 105 exchanges · 12 tasks
AGPL-3.0
Sentry 26.9.0
error tracking and issue triage (application monitoring)
34 tools · 72 exchanges · 12 tasks
FSL-1.1-Apache-2.0
HashiCorp Vault 1.21.4
secrets and identity management (configuration store, policies, identity)
43 tools · 85 exchanges · 12 tasks
BUSL-1.1
HashiCorp Nomad 1.11.3
workload orchestration (cluster scheduler)
39 tools · 131 exchanges · 12 tasks
BUSL-1.1
Redpanda Console 3.12.0
event streaming (Kafka-compatible topics, consumer groups and schema registry)
37 tools · 87 exchanges · 8 tasks
BSL-1.1
Airbyte 2.4.0
data integration pipelines (ELT connectors, connections and sync jobs)
25 tools · 64 exchanges · 11 tasks
Elastic License 2.0 (ELv2)
n8n 1.123.85
workflow automation
30 tools · 72 exchanges · 9 tasks
Sustainable Use License 1.0
Keycloak 26.8.1
identity and access management (IAM)
40 tools · 73 exchanges · 12 tasks
Apache-2.0
Metabase 0.64.2.8
business intelligence and dashboards
42 tools · 120 exchanges · 12 tasks
AGPL-3.0
Umami 3.4
privacy-focused web analytics
49 tools · 100 exchanges · 12 tasks
MIT
NocoDB 2026.09.1
no-code databases (spreadsheet-style tables)
28 tools · 105 exchanges · 12 tasks
AGPL-3.0
Baserow 2.4.1
no-code databases (spreadsheet-style tables)
31 tools · 83 exchanges · 12 tasks
MIT
Grist 1.7.21
spreadsheet-database (workspaces, documents, tables, records)
31 tools · 65 exchanges · 12 tasks
Apache-2.0
Directus 11.17.4
headless CMS and data platform
43 tools · 90 exchanges · 12 tasks
BSL-1.1
Mattermost 12.0.0
team chat (channels, posts, teams)
37 tools · 129 exchanges · 12 tasks
Team Edition, MIT-licensed server core
Ghost 5.130
publishing and newsletters (CMS)
36 tools · 76 exchanges · 12 tasks
MIT
BookStack 26.09.1
team wiki and documentation (shelves, books, chapters, pages)
31 tools · 65 exchanges · 12 tasks
MIT
listmonk 6.2
newsletters and mailing lists
29 tools · 68 exchanges · 12 tasks
AGPL-3.0
Vikunja 2.7
task and project management (to-do lists, kanban)
28 tools · 102 exchanges · 12 tasks
AGPL-3.0-or-later
EspoCRM 10.0.10
customer relationship management (CRM)
88 tools · 131 exchanges · 12 tasks
AGPL-3.0
Paperless-ngx 2.20
document management and archiving
41 tools · 108 exchanges · 12 tasks
GPL-3.0
Memos 0.31.0
note-taking and microblogging (memos)
26 tools · 67 exchanges · 12 tasks
MIT
linkding 1.47.0
self-hosted bookmark manager
23 tools · 64 exchanges · 12 tasks
MIT
Shlink 5.1.7
URL shortening and visit analytics
17 tools · 66 exchanges · 7 tasks
MIT
Miniflux 2.3.3
feed reader (RSS and Atom)
22 tools · 86 exchanges · 7 tasks
Apache-2.0
PhotoPrism 261007
photo library and media management
13 tools · 277 exchanges · 10 tasks
AGPL-3.0
Lychee 6.10.4
photo management (self-hosted photo gallery)
15 tools · 134 exchanges · 12 tasks
MIT
Pinry 2.1.13
visual pin boards (tiling image bookmarks with boards and tags)
12 tools · 568 exchanges · 6 tasks
BSD-2-Clause
Immich 3.3.3
self-hosted photo and video library
45 tools · 120 exchanges · 12 tasks
AGPL-3.0
Jellyfin 10.10.7
media server and personal media library
34 tools · 72 exchanges · 12 tasks
GPL-2.0
Navidrome 0.64.2
self-hosted music streaming server
25 tools · 71 exchanges · 12 tasks
GPL-3.0
Komga 1.29.1
comics and ebook library server
42 tools · 116 exchanges · 12 tasks
MIT
Sonarr 4.0.20.3014
TV library management (profiles, custom formats, tags, connections)
48 tools · 98 exchanges · 12 tasks
GPL-3.0
Homebox 0.27.0
home inventory and asset tracking
28 tools · 91 exchanges · 10 tasks
AGPL-3.0
Grocy 4.7.1
household groceries, stock and chores (ERP)
45 tools · 75 exchanges · 12 tasks
MIT
Mealie 3.28
recipe management and meal planning
39 tools · 139 exchanges · 12 tasks
AGPL-3.0
Art Institute of Chicago API v1
art museum collection (artworks, artists, galleries, exhibitions)
11 tools · 26 exchanges · 12 tasks
CC0 1.0 (collection data)
Library of Congress JSON API (photos) v1
historical photograph collections (public API)
11 tools · 29 exchanges · 12 tasks
U.S. public domain
openFDA v1
drug, device and food recalls, drug product listings and device classifications (U.S. government public API)
9 tools · 53 exchanges · 12 tasks
U.S. public domain
USGS Earthquake Catalog 2.8.1
earthquake catalog and real-time seismic event feeds (geoscience)
10 tools · 53 exchanges · 7 tasks
U.S. public domain
NOAA Tides and Currents API v1
coastal water levels, tide predictions and tide-station metadata (public data API)
12 tools · 56 exchanges · 9 tasks
U.S. public domain
Federal Register API v1
U.S. federal rules, proposed rules, notices and presidential documents
10 tools · 52 exchanges · 10 tasks
U.S. public domain
Treasury Fiscal Data API v1/v2
federal finance (public debt, exchange rates, Treasury statements, auctions and agency receivables)
16 tools · 52 exchanges · 12 tasks
U.S. public domain
Healthchecks 4.4
Cron job and heartbeat monitoring
18 tools · 69 exchanges · 12 tasks
BSD-3-Clause
Mailpit 1.31.6
Email testing (SMTP capture and web mailbox)
15 tools · 68 exchanges · 11 tasks
MIT
NetBox 4.7.3
Network source of truth (DCIM and IPAM)
42 tools · 100 exchanges · 12 tasks
Apache-2.0
WordPress 7.1.3
Blogging and content management (CMS)
21 tools · 64 exchanges · 12 tasks
GPL-2.0-or-later
Tandoor Recipes 2.6.15
Recipe management and meal planning
36 tools · 130 exchanges · 12 tasks
AGPL-3.0 with Commons Clause
PocketBase 0.40.5
Backend-as-a-service database and admin dashboard
25 tools · 86 exchanges · 12 tasks
MIT
Run one
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/PinryEnv-REA --bundle pinry --task smoke # Docker, no model
agent-env hf run earakely-scale/PinryEnv-REA --bundle pinry --eval pinry-retrieval --model openai/gpt-oss-120b:groqWhat replay means, and does not
- A tool call returns the recorded response whose path and query values best match the arguments;
replay.exact_matchsays whether it matched exactly. - There is no state: a write returns its one recorded answer and changes nothing, so retrieval tasks are hard-gradable and write tasks are weak.
- Self-hosted apps ran on throwaway machines over invented data (people, places, projects); public-API environments hold real public records frozen on their capture date.
- 5 environments were captured by REA's own
capture-browser-scenario: REA drove Chromium, redacted credential headers and declared secrets, and the manifest read that capture directly with no HAR or scrubbing step. The rest were captured as HARs and inspected by REA. - Each manifest states what replay cannot know (authorization model, response semantics beyond the recorded exchanges, state transitions).