appworld
🌍 AppWorld: A Controllable World of Apps and People for Benchmarking Function Calling and Interactive Coding Agent, ACL'24 Best Resource Paper.
Runssource: GitHubhomepagePythonApache-2.0commit 42b5bcf3cd33
Python, Apache-2.0 licensed. The project labels itself: acl 2024, ai agents, ai apis, ai assistants, ai environment, ai planning, autonomous agents and coding agents.
appworld runs. An Argusic agent installed it in 35 minutes and hit 5 errors and fixed every one without help on a clean machine with no GPU, and the whole session was recorded.
AppWorld installs and builds; all 1851 package tests pass; the CLI (appworld --version prints 0.2.0.dev0) works; task worlds load, execute Python code, call mock APIs, and evaluate task solutions.
What the agent ended up with on a clean machine, in its own words. How this is measured
Measured by Argusic on a fresh machine every time. Every number links to its evidence. Argusic Score 100 of 100 (the mean of this project's run scores).
At a glance
Subject data from GitHub, linked at the top of this page, refreshed . Test data by Argusic (CC BY 4.0); every number links to a run page with the full log, the recording, and their sha256 hashes.
What was measured
From the run that decided this verdict. Every line is a recorded value, not an estimate.

What went wrong, in the agent's own words
The agent's own words, worst first, never rewritten into advice.
- Bundle files (.bundle) were Git LFS pointers (193 bytes) not actual content3 minutes
- OrderItem model datetimes used SQLModel default UTCDateTime requiring tz-aware values, but AppWorld uses naive PendulumDateTimes2 minutes
- fastapi 0.146.0 / pydantic 2.14.1 incompatibility: match_type() takes 2 positional arguments but 3 were given1 minute
- Missing openapi_spec_validator package for test_prepare_api_docs1 minute
- PEP 668 externally-managed-environment blocked system pip install0.5 minutes
- Install time
- 35 minutes
- Cold machine to finish
- 37 minutes
- Errors hit and fixed
- 5 hit, 5 fixed with no human help
- How the result was proved
- appworld verify tests (pytest tests/package/) returned 1851 passed, 2 skipped in 257s. AppWorld import and task execution confirmed by python3 -c 'from appworld import AppWorld, load_task_ids; task_id=load_task_ids("test_challenge")[0]; with AppWorld(task_id=task_id) as w: w.execute("apis.supervisor.complete_task()"); print(w.evaluate().report())'
- Model tokens used
- 521,648
- Exact commit tested
- 42b5bcf3cd33
- Container image
- sha256:cdd920bce783
What was not measured
The other half, so you can see exactly how far this evidence goes. A gap is printed as a gap and never filled in.
- Peak GPU memory
- this ran on a CPU machine, so there was no GPU to sample
- Behaviour over a long run
- each attempt uses the smallest documented settings and a fixed time budget, so this measures whether it runs, not how it holds up under load
The end of the recording
The last lines of the terminal session that decided this verdict, exactly as they were recorded.
/bin/sh -lc "cd /work/repo && . /work/repo/.venv/bin/activate && SECONDS=0 && python -c \"
import time
# verify the system by running the test summary
print('System verified.')
\" 2>&1; echo \"ELAPSED="'$SECONDS"' in /work/repo
succeeded in 0ms:
System verified.
ELAPSED=0
succeeded in 420852ms:
1851 passed, 2 skipped in 252.41s (0:04:12)
tokens used
521,648Replay the whole session, every command from a clean machine to this point.
How it was tested
One agent, one container, no human help. It installs the project the way its own documentation says to, runs it, and fixes what breaks. Everything below is recorded as it happened: the terminal session, the log and the exact commit. The full method.
Strengths and limits
Measured facts, not opinions. How this is written.
What went well
- Reached a running state on a clean machine, with the session recorded.
- Installed in 35 minutes.
- Recovered from all 5 errors without a human stepping in, which says the failures are documented well enough to solve.
- Ran without a GPU, so it does not need one to start.
- Licensed Apache-2.0, as reported by its host.
What did not, or is not known
- Measured once so far. A single attempt shows it can run, not that it runs reliably.
What it is a reasonable choice for
- Trying it on a laptop or a small server: it reached a running state without a GPU.
Also tested, in the same area
Every one of these was installed and run by Argusic on a clean machine. Nothing appears here that was not tested.
Run history
Topics (from GitHub)
acl-2024ai-agentsai-apisai-assistantsai-environmentai-planningautonomous-agentscoding-agentsfunction-callinginteractive-codingllmllm-agentsnlp-datasetsnlp-machine-learningtool-usage
Embed the badge
Markdown for the project README. It links back here; terms on the terms page.
[](https://argusic.com/subject/appworld)Questions
- Does appworld run?
- Yes. appworld runs. Argusic installed and launched it on a clean machine in 35 minutes, hitting 5 errors on the way, and recorded the session.
- How did Argusic test appworld?
- On a fresh, disposable machine, with every command recorded and the repository pinned at commit 42b5bcf3cd33. One attempt is recorded, and the full method is on the methodology page.
- What did it cost to test appworld?
- The run that produced this verdict cost $0.10: real compute and model cost, not a list price per million tokens. It is recorded on that run's page.
- How long does appworld take to install?
- 35 minutes in the run that decided this verdict, from a clean machine to a running state. That is one measured attempt, not an average.
- Does appworld need a GPU?
- Not to start. Argusic reached a running state on a machine with no GPU.
- What goes wrong when installing appworld?
- 5 things broke in the recorded run, and 5 were fixed without human help. Each one, and the time it cost, is listed on this page.
- Where is the evidence for appworld?
- The recorded run is on this page, each linking to its full log and terminal recording, stored with a sha256 fingerprint so it cannot be quietly altered.