ai-research 0.1.25

Python SDK for the AIR Backend API

AIR SDK

The AIR SDK is a Python client for the AIR platform — a suite of AI assistant tools for scientific research. It gives you programmatic access to AI-assisted pipeline steps (idea generation, literature search, methods, data analysis, paper writing, peer review) as well as a set of standalone tools you can use independently.

Key capabilities:

• AI-assisted research pipeline — use AI assistants for idea generation, literature search, methods development, data analysis, paper writing, and review as a connected workflow or individual steps
• Local code execution — the results, one_shot, and deep_research methods coordinate an AI agent on the backend while all generated code runs locally on your machine in an isolated virtual environment
• File sync — push local data files to a project and pull generated outputs (plots, LaTeX, reports) back to disk after each step
• Standalone AI tools — extract keywords, download and summarise arXiv papers, enhance text with paper context, run OCR, or get an AI review of a paper PDF independently of any project

Installation

Install the SDK from PyPI:

pip install air-research

Quick Start

import air

client = air.AIR(api_key="air_k1_...", base_url="http://localhost:8000", local_dir="~/my-research")

# Standalone tools
keywords = client.keywords("dark matter and lensing", n=5, kw_type="aas")
enhanced = client.enhance("My research with https://arxiv.org/abs/2301.12345")

# Full research workflow
project = client.create_project("my-research", data_description="We study...")
idea = project.idea()
literature = project.literature()
methods = project.methods()
results = project.results()
paper = project.paper(journal="AAS")
review = project.review()

# File access
print(project.get_file("Iteration0/input_files/idea.md"))
print(project.list_files())

# Sync files locally
project.pull_files()                       # download all project files
project.push_files("local/data.csv")       # upload a local file

# One-shot and deep-research tasks
result = client.one_shot("Analyse the attached dataset and produce a summary plot")
result = client.deep_research("Investigate the impact of dark energy on large-scale structure")

Configuration

Set environment variables, or pass them directly to the constructor:

Variable	Description
AIR_API_KEY	API key (air_k1_...)
AIR_BASE_URL	Backend URL (default: http://localhost:8000)
AIR_LOCAL_DIR	Local root for pull_files / auto-pull (default: ~/ai-scientist)
AIR_WORK_DIR	Local root for results, one_shot, deep_research code execution (default: ~/ai-scientist)

client = air.AIR(
    api_key="air_k1_...",
    base_url="https://api.example.com",
    local_dir="~/my-research",   # where pull_files writes files
)

API Reference

AIR class

Method	Description
keywords(text, n=5, kw_type="unesco")	Extract keywords
arxiv(text)	Download arXiv papers from URLs in text
enhance(text, max_workers=2, max_depth=10)	Enhance text with arXiv context
ocr(file_path)	Process PDF with OCR (server path)
review(file_path, *, thoroughness, figures_review, verify_statements, review_maths, review_numerics, emails, timeout)	Standalone paper review with Skepthical
create_project(name, data_description, iteration, local_dir, auto_pull=True)	Create a project
get_project(name, auto_pull=True)	Get existing project
list_projects()	List all projects
delete_project(name)	Delete a project
one_shot(task, *, model, max_rounds, agent, work_dir, timeout, on_output, python_path, venv_path)	Run a one-shot task with local code execution
deep_research(task, *, engineer_model, researcher_model, planner_model, max_plan_steps, work_dir, timeout, on_output, ...)	Run a multi-step deep-research task with local code execution

Project class

Method	Description
idea(mode="fast", iteration=0, timeout=600)	Generate research idea
literature(iteration=0, timeout=600)	Run literature search
methods(mode="fast", iteration=0, timeout=600)	Develop methods
results(iteration=0, timeout=86400, max_attempts=10, max_steps=6, work_dir, on_output, python_path, venv_path)	Run analysis/experiments with local code execution; plots are auto-uploaded
paper(journal="NONE", iteration=0, timeout=900)	Write paper
review(iteration=0, timeout=600, **kwargs)	Run review (kwargs: review_engine, review_thoroughness, pdf_version, emails, etc.)
get_file(path)	Read a project file
list_files()	List all project files
write_file(path, content, encoding="utf-8")	Write a file
pull_files(local_dir, iteration=0, key_only=False)	Download project files to a local directory
push_files(local_path, remote_path=None)	Upload a local file or directory to the project
delete()	Delete the project

auto_pull

When auto_pull=True (default), key project files are automatically downloaded to local_dir after each pipeline step (idea, methods, results, paper). Set auto_pull=False to disable this behaviour.

Standalone review() on AIR

result = client.review(
    "/server/path/to/paper.pdf",
    thoroughness="High",       # "Standard" (default) or "High"
    figures_review=True,
    verify_statements=True,
    review_maths=False,
    review_numerics=False,
    emails=["[e-mail]"],
)

one_shot and deep_research

Both methods connect via WebSocket and execute generated code locally on your machine:

result = client.one_shot(
    "Plot a histogram of the attached CSV",
    model="gpt-4.1-2025-04-14",
    work_dir="~/my-tasks",
)
print(result.output)   # streamed text
print(result.files)    # list of created local files

result = client.deep_research(
    "Reproduce Figure 3 from arxiv:2301.12345",
    max_plan_steps=5,
    work_dir="~/my-tasks",
)

Docs

Install the documentation with:

pip install air-research[docs]

and build it with

mkdocs serve --livereload

Tests

Test with pytest:

pytest tests

You can also skip slow tests (e.g. idea generation) with

pytest tests/ -m "not slow".