system1-agents/pyproject.toml at main · ThinkFlowLab/system1-agents · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
[project]
name = "system1-agents"
version = "0.1.0"
description = "System 1 decision models (TypeSafe Jev, Laya, Cua-S1) in the model slot of openJiuwen agents, with browser, tool and rail agents and evals against a chat model on the same agent."
readme = "README.md"
license = "Apache-2.0"
license-files = ["LICENSE", "NOTICE", "THIRD_PARTY_LICENSES.md"]
authors = [{ name = "ThinkFlowLab" }]
requires-python = ">=3.11"
dependencies = [
    # Tag on the jiuwen-jev branch of the ThinkFlowLab fork; CONTRIBUTING.md § The openjiuwen pin lists what it adds over upstream develop.
    "openjiuwen @ git+https://github.com/ThinkFlowLab/agent-core@jj-0.1.0",
    # Imported directly; openjiuwen depends on all three, so they add nothing to the install.
    "httpx>=0.28",
    "python-dotenv>=1.0",
    "mcp>=1.26",
]

[project.urls]
Homepage = "https://github.com/ThinkFlowLab/system1-agents"
Repository = "https://github.com/ThinkFlowLab/system1-agents"
Issues = "https://github.com/ThinkFlowLab/system1-agents/issues"

[project.scripts]
s1a = "s1a.entry:s1a"
s1a-mcp = "s1a.entry:s1a_mcp"

[project.optional-dependencies]
blackjack = ["rlcard>=1.2"]
alfworld = ["alfworld>=0.4", "textworld>=1.7"]  # plus ALFWORLD_DATA and Python 3.11; see evals/README.md
# evals/replay/thor_replay.py: the AI2-THOR scene behind an ALFWorld trial. ai2thor 2.1.0 is the 2019 Unity build the games
# were generated in (x86_64 on macOS, Rosetta on Apple silicon; downloaded on first run). werkzeug 2.0.3 is alfworld's own
# pin for that build's server; numpy and opencv float with the base install.
alfworld-visual = [
    "system1-agents[alfworld]",
    "ai2thor==2.1.0",
    "opencv-python-headless>=4.10",
    "werkzeug==2.0.3",
    "torch>=2",
    "torchvision>=0.15",
]
report = ["pillow>=10", "playwright>=1.45"]  # evals.replay: pages, GIFs; never imported by a runner
laya = ["laya>=0.3.4"]  # the in-process decision model behind --model laya; pulls torch and transformers
cua = [  # Cua-S1 Nano behind --model cua; pinned to the Cua PR that ships the checkpoints (trycua/cua#4023), pulls torch
    "cua-s1 @ git+https://github.com/trycua/cua.git@aea61b6eb97e2d8c0f6f71eb804e5769fe910af4#subdirectory=libs/cua-s1/python",
    "huggingface-hub>=0.24",
]
dev = ["pytest>=8", "pytest-asyncio>=0.24", "ruff>=0.6", "ty>=0.0.83"]

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[tool.hatch.metadata]
allow-direct-references = true  # the openjiuwen git pin above

[tool.hatch.build.targets.wheel]
packages = ["s1a"]

[tool.hatch.build.targets.sdist]
only-include = ["s1a", "README.md", "LICENSE", "NOTICE", "THIRD_PARTY_LICENSES.md"]

[tool.pytest.ini_options]
asyncio_default_fixture_loop_scope = "function"
testpaths = ["tests"]
pythonpath = ["."]
addopts = ["--ignore=tests/system"]  # tests/system launches a browser through @playwright/mcp (Node)

[tool.ty.src]
include = ["s1a"]

[[tool.ty.overrides]]
# The game agents and the in-process decision models import their extras; the core install has none of them.
include = [
    "s1a/agents/alfworld.py",
    "s1a/agents/blackjack.py",
    "s1a/decision_models/cua.py",
    "s1a/decision_models/laya.py",
]

[tool.ty.overrides.rules]
unresolved-import = "ignore"

[tool.ruff]
line-length = 120
target-version = "py311"

[tool.ruff.lint]
select = ["E4", "E7", "E9", "F"]  # the rule set the code was written against; ruff 0.16 enables many more by default