struai 2.5.0

StruAI Drawing Analysis SDK - AI-powered construction drawing analysis

StruAI SDK (Python + JavaScript)

Official SDKs for the StruAI Drawing Analysis API.

• Python package: struai (PyPI)
• JavaScript package: struai (npm, source in js/)
• Endpoint + response-shape reference: docs/HTTP_ENDPOINTS.md

Install

pip install struai
npm install struai

Environment

export STRUAI_API_KEY=your_api_key
# Optional: defaults to https://api.stru.ai (SDK appends /v1 automatically)
export STRUAI_BASE_URL=https://api.stru.ai

Python Quick Start

import os
from struai import StruAI

client = StruAI(api_key=os.environ["STRUAI_API_KEY"])

# Tier 1: drawings
result = client.drawings.analyze("structural.pdf", page=12)
print(result.id, result.processing_ms)

# Tier 2: projects + docquery
project = client.projects.create(name="Building A", description="Structural set")
job = project.sheets.add(page=12, file_hash=client.drawings.compute_file_hash("structural.pdf"))
sheet = job.wait(timeout=180)

hits = project.docquery.search("beam connection", limit=5)
print(len(hits.hits))

# Tier 3: reviews (default review team)
review = client.reviews.create(
    file_hash=client.drawings.compute_file_hash("structural.pdf"),
    pages="12,13",
    project_ids=[project.id],
    custom_instructions="Focus on cross-sheet coordination.",
)
final_review = review.wait(timeout=900, poll_interval=5)
print(final_review.status, len(review.issues()))

# Tier 3: reviews (custom scout + custom specialist team)
custom_review = client.reviews.create(
    file_hash=client.drawings.compute_file_hash("structural.pdf"),
    pages="13",
    project_ids=[project.id],
    scout="Look broadly for issues worth deeper investigation.",
    specialists_common="All specialists should stay grounded in DocQuery/search-docs evidence.",
    specialists=[
        {"name": "Cross Reference Checker", "instructions": "Trace all detail, section, and sheet references."},
        {"name": "Constructability Reviewer", "instructions": "Focus on missing dimensions and buildability gaps."},
    ],
)
print(custom_review.id)

# Direct upload path also works
upload_review = client.reviews.create(
    file="structural.pdf",
    pages=12,
)
print(upload_review.id)

Real Workflow Examples

Python examples (/examples):

# Drawings-only flow (hash, cache probe, analyze)
python3 examples/test_prod_page12.py --pdf /absolute/path/to/structural.pdf --page 12

# Full projects + docquery workflow
python3 examples/test_prod_page12_full.py --pdf /absolute/path/to/structural.pdf --page 12

# Full workflow + crop demo
python3 examples/test_prod_page12_full.py \
  --pdf /absolute/path/to/structural.pdf --page 12 \
  --crop-output /absolute/path/to/crop.png

# Optional cleanup after full workflow
python3 examples/test_prod_page12_full.py --pdf /absolute/path/to/structural.pdf --cleanup

# Async workflow
python3 examples/async_projects_workflow.py --pdf /absolute/path/to/structural.pdf --page 12

# Review workflow (start + refresh)
python3 examples/review_workflow.py --file-hash your_file_hash --pages 13

# Review workflow (wait for terminal status)
python3 examples/review_workflow.py --file-hash your_file_hash --pages 13 --wait

# Page-13 review cookbook (default review team)
python3 examples/page13_review_cookbook.py

# Page-13 review cookbook (custom scout + custom specialist team)
STRUAI_REVIEW_SCOUT_FILE=/absolute/path/to/examples/prompts/page13_review/scout.md \
STRUAI_REVIEW_SPECIALISTS_COMMON_FILE=/absolute/path/to/examples/prompts/page13_review/specialists_common.md \
STRUAI_REVIEW_SPECIALISTS_FILE=/absolute/path/to/examples/prompts/page13_review/specialists.json \
python3 examples/page13_review_cookbook.py

Page-12 cookbook with all 10 operations (including cypher and crop):

• examples/PAGE12_COOKBOOK.md
• examples/prompts/page13_review/README.md
• examples/REVIEWS_QUICKSTART.md

JavaScript examples (/js/scripts):

cd js
npm install
npm run build

# Drawings-only flow
STRUAI_API_KEY=... STRUAI_BASE_URL=https://api.stru.ai \
STRUAI_PDF=/absolute/path/to/structural.pdf STRUAI_PAGE=12 \
node scripts/drawings_quickstart.mjs

# Full projects + docquery workflow
STRUAI_API_KEY=... STRUAI_BASE_URL=https://api.stru.ai \
STRUAI_PDF=/absolute/path/to/structural.pdf STRUAI_PAGE=12 \
node scripts/projects_workflow.mjs

# Full workflow + crop demo
STRUAI_API_KEY=... STRUAI_BASE_URL=https://api.stru.ai \
STRUAI_PDF=/absolute/path/to/structural.pdf STRUAI_PAGE=12 \
STRUAI_CROP_OUTPUT=/absolute/path/to/crop.png \
node scripts/projects_workflow.mjs

# Review workflow (default review team)
STRUAI_API_KEY=... STRUAI_BASE_URL=https://api.stru.ai \
STRUAI_REVIEW_FILE_HASH=your_file_hash STRUAI_REVIEW_PAGES=13 \
node scripts/reviews_workflow.mjs

# Review workflow (custom scout + custom specialist team)
STRUAI_API_KEY=... STRUAI_BASE_URL=https://api.stru.ai \
STRUAI_REVIEW_FILE_HASH=your_file_hash STRUAI_REVIEW_PAGES=13 \
STRUAI_REVIEW_SCOUT_FILE=/absolute/path/to/examples/prompts/page13_review/scout.md \
STRUAI_REVIEW_SPECIALISTS_COMMON_FILE=/absolute/path/to/examples/prompts/page13_review/specialists_common.md \
STRUAI_REVIEW_SPECIALISTS_FILE=/absolute/path/to/examples/prompts/page13_review/specialists.json \
node scripts/reviews_workflow.mjs

Python API Reference

Async API (AsyncStruAI) mirrors the same resource shape and method names; use await.

Client

• StruAI(api_key=None, base_url="https://api.stru.ai", timeout=60, max_retries=2)
• AsyncStruAI(api_key=None, base_url="https://api.stru.ai", timeout=60, max_retries=2)
• client.drawings
• client.projects
• client.reviews

Drawings (client.drawings)

• analyze(file=None, page=1, file_hash=None) -> DrawingResult
• check_cache(file_hash) -> DrawingCacheStatus
• compute_file_hash(file) -> str

Projects Top-Level (client.projects)

• create(name, description=None) -> ProjectInstance
• list() -> list[Project]
• open(project_id, name=None, description=None) -> ProjectInstance
• delete(project_id) -> ProjectDeleteResult

Reviews Top-Level (client.reviews)

• create(file=None, pages=1|"1,3,5-7"|"all", file_hash=None, project_ids=None, scout=None, specialists_common=None, specialists=None, custom_instructions=None) -> ReviewInstance
  • Pass exactly one of file or file_hash.
  • Raises ValueError if both are missing or both are provided.
  • specialists must be a non-empty list of {name, instructions} objects when provided.
  • Omit scout, specialists_common, and specialists to use the default review team.
• list(status=None) -> list[Review]
• get(review_id) -> ReviewInstance
• open(review_id) -> ReviewInstance

Project Instance (project)

Properties:

• id, name, description, data
• sheets, docquery

Methods:

• delete() -> ProjectDeleteResult

Review Instance (review)

Properties:

• id, data

Methods:

• refresh() -> Review
• status() -> Review
• wait(timeout=900, poll_interval=5) -> Review
  • Raises ReviewFailedError if the review reaches failed.
  • Raises TimeoutError if the timeout elapses first.
• questions() -> list[ReviewQuestion]
• issues() -> list[ReviewIssue]

Sheets (project.sheets)

• add(file=None, page=1|"1,3,5-7"|"all", file_hash=None, source_description=None, on_sheet_exists=None, community_update_mode=None, semantic_index_update_mode=None) -> Job | JobBatch
• delete(sheet_id) -> SheetDeleteResult
• job(job_id, page=None) -> Job

DocQuery (project.docquery)

• node_get(uuid) -> DocQueryNodeGetResult
• sheet_entities(sheet_id, entity_type=None, limit=200) -> DocQuerySheetEntitiesResult
• search(query, index="entity_search", limit=20) -> DocQuerySearchResult
• neighbors(uuid, mode="both", direction="both", relationship_type=None, radius=200.0, limit=50) -> DocQueryNeighborsResult
• cypher(query, params=None, max_rows=500) -> DocQueryCypherResult
• sheet_summary(sheet_id, orphan_limit=10) -> DocQuerySheetSummaryResult
• sheet_list() -> DocQuerySheetListResult
• reference_resolve(uuid, limit=100) -> DocQueryReferenceResolveResult
• crop(output, uuid=None, bbox=None, page_hash=None) -> DocQueryCropResult

CLI parity: project-list maps to client.projects.list(), and the remaining 9 commands map to project.docquery.*, for full 10-command parity.

Python cypher + crop example:

project = client.projects.open("proj_86c0f02e")
rows = project.docquery.cypher(
    "MATCH (n:Entity {project_id:$project_id}) RETURN count(n) AS total",
    params={},
    max_rows=1,
)

crop = project.docquery.crop(
    uuid="entity-uuid-here",
    output="/absolute/path/to/crop.png",
)
print(rows.records[0]["total"], crop.output_path, crop.bytes_written)

Jobs

Job (single-page ingest result):

• id, page
• status() -> JobStatus
• wait(timeout=120, poll_interval=2) -> SheetResult

JobBatch (multi-page ingest result):

• jobs, ids
• status_all() -> list[JobStatus]
• wait_all(timeout_per_job=120, poll_interval=2) -> list[SheetResult]

Reviews

Review:

• review_id, status, total_pages, pages, progress
• is_running, is_complete, is_partial, is_failed, is_terminal
• Status values: running, completed, completed_partial, failed
• pages and total_pages are populated by POST /v1/reviews; later refresh() / get() calls reflect the slimmer GET /v1/reviews/{review_id} payload and may omit them.
• progress.specialist.active contains live in-progress specialist rows with question_id, agent, turns_used, max_turns, and updated_at when the server exposes them.

ReviewQuestion:

• Includes raw_model_output, which is preserved as nested JSON when present.

HTTP Endpoints Covered

Full endpoint + response-shape reference: docs/HTTP_ENDPOINTS.md

Tier 1:

• POST /v1/drawings
• GET /v1/drawings/cache/{file_hash}

Tier 2:

• POST /v1/projects
• GET /v1/projects
• DELETE /v1/projects/{project_id}
• POST /v1/projects/{project_id}/sheets
• DELETE /v1/projects/{project_id}/sheets/{sheet_id}
• GET /v1/projects/{project_id}/jobs/{job_id}
• GET /v1/projects/{project_id}/node-get
• GET /v1/projects/{project_id}/sheet-entities
• GET /v1/projects/{project_id}/search
• GET /v1/projects/{project_id}/neighbors
• POST /v1/projects/{project_id}/cypher
• POST /v1/projects/{project_id}/crop

Tier 3:

• POST /v1/reviews
• GET /v1/reviews
• GET /v1/reviews/{review_id}
• GET /v1/reviews/{review_id}/questions
• GET /v1/reviews/{review_id}/issues

JavaScript Reference

See js/README.md for complete JS method signatures and usage patterns.

License

MIT