acquivela

AcquiVela

Document Acquisition Framework — Frozen Architecture v0.1

Note: AcquiVela v0.1.0 is a pre-1.0 release. Some internal orchestration APIs (Pipeline, Registry) are exposed for use but are not frozen public contracts and may change. See Governance below.

What is AcquiVela?

AcquiVela is a capability-oriented document acquisition framework that orchestrates specialized acquisition engines (Docling, MinerU) through frozen public contracts and a versioned Canonical Document Model (CDM).

Scope: Document acquisition only (PDF, DOCX, scanned images, and more). Non-goals: Embeddings, retrieval, vector DB, chunking, LLM reasoning, knowledge extraction.

Governed Public Contracts

The following are frozen per ADR-0001 through ADR-0006 and carry semantic-versioning compatibility guarantees:

Internal Orchestration Infrastructure

The following components are internal and not frozen public contracts:

Plugins may NOT import Pipeline, Registry, or public_result — enforced via import-linter (ADR-0004 §3.2, ADR-0006 §3.3).

Installation

# Base installation (no acquisition engines)
pip install acquivela

# With Docling support (PDF, DOCX, images, and more)
pip install acquivela[docling]

# With MinerU support (high-accuracy PDF)
pip install acquivela[mineru]

# With both engines
pip install acquivela[docling,mineru]

System Dependencies

Some formats require additional system-level tools:

• Legacy Office formats (.doc, .ppt, .xls): Requires LibreOffice installed and on PATH
• OpenDocument formats (.odt, .ods, .odp): Requires odfdo Python package (included with docling extras)

Quick Start

# Recommended: use the M15 stable facade
from acquivela import AcquireRequest, acquire

result = acquire(
    AcquireRequest(file_path="document.pdf")
)

if result.status == "success":
    doc = result.canonical_document
    for page in doc.pages:
        for block in page.text_blocks:
            print(block.text)

Using the M15 facade (recommended normal path)

from acquivela import AcquireRequest, acquire

result = acquire(
    AcquireRequest(file_path="document.pdf")
)

if result.status == "success":
    doc = result.canonical_document
    for page in doc.pages:
        for block in page.text_blocks:
            print(block.text)

Using Engine Hints

# Force a specific engine
request = AcquireRequest(
    file_path="document.pdf",
    engine_hint="docling"
)

Using Capability Filtering

from acquivela.capabilities import Capabilities

# Require text extraction capability
request = AcquireRequest(
    file_path="document.pdf",
    required_capability=Capabilities.DOCUMENT_TEXT
)

Known Limitations

MinerU

MinerU has a known upstream issue (PPDocLayoutV2Config missing reading_order_config) that causes acquisition failures. AcquiVela correctly contains this failure — the pipeline returns status="failure" without exceptions escaping. The MinerU plugin reports healthy at the health-check level because it is importable, but actual acquisition will fail until the upstream issue is resolved.

Legacy Office Formats

.doc, .ppt, and .xls files require LibreOffice to be installed and available on PATH. Without LibreOffice, Docling cannot convert these legacy formats to modern equivalents.

OpenDocument Formats

.odt, .ods, and .odp files require the odfdo Python package. This is included when installing with pip install acquivela[docling].

XML and JSON

Generic XML files are not supported. Docling supports specific XML dialects only (USPTO, JATS, XBRL, DOCLANG). Similarly, only json_docling format is supported for JSON input.

CDM v0.2 Exclusions

The Canonical Document Model (CDM) v0.2 captures text content only. Tables, images, equations, bounding boxes, and rich provenance are intentionally excluded from the current model.

Architecture

Package Structure

acquivela/
├── __init__.py          # Public exports + internal orchestration (see Governance)
├── cdm.py               # CanonicalDocument, Page, TextBlock (PUBLIC/STABLE)
├── models.py            # AcquireRequest (PUBLIC/STABLE)
├── plugin.py            # Plugin SPI, PluginHealth (PUBLIC/STABLE)
├── result.py            # PluginResult (INTERNAL - not exported at root)
├── public_result.py     # AcquireResult, Diagnostics (PUBLIC/STABLE)
├── capabilities.py      # Capabilities (PUBLIC/STABLE)
├── registry.py          # Registry (INTERNAL - may change)
├── pipeline.py          # Pipeline (INTERNAL - may change)
├── execution_context.py # ExecutionContext (INTERNAL)
└── plugins/             # Plugin implementations (one class per module)

Status

• Charter: FROZEN
• ADR-0001 Project Boundaries: FROZEN
• ADR-0002 Execution Model: FROZEN
• ADR-0003 Core Data Contracts: FROZEN
• ADR-0004 Plugin & Execution Strategy: FROZEN
• ADR-0005 Observability & Diagnostics: FROZEN
• ADR-0006 Registry & Plugin Discovery: FROZEN

Development

# Install with dev dependencies
pip install -e .[dev]

# Run contract tests
pytest tests/test_contracts.py

# Run import-linter validation
pytest tests/test_import_linter.py

# Or run import-linter directly
importlinter lint

License

MIT License. See LICENSE for details.