fileroute

Fileroute describes where project artifacts come from, where they live locally, and where they should be published. A small YAML or JSON catalog connects local files with SharePoint, Google Drive, and S3 URLs, so a document or data export can keep its provenance and multiple destinations in one place. The CLI can preview changes without credentials, visualize descriptor workflows as SVG, HTML, Markdown, or Mermaid, pull remote inputs, and push supported outputs; the same descriptor, graph, and transfer APIs are available from Python. Use it inside a project for repeatable, versioned workflows, or run it as a standalone tool to inspect and retrieve individual files. Python 3.11 or newer is required.

Choose how to run it

Within a project: add Fileroute to that project's dependencies, commit its uv.lock and descriptor, and run commands from the project root. uv run uses the project's environment and resolves its declared dependencies before running the command. This is the better fit for automation and Python API imports.

uv add fileroute
uv run fileroute diagram config/fileroute.yaml
uv run fileroute pull config/fileroute.yaml --dry-run
uv run fileroute push config/fileroute.yaml --dry-run

Use uv add 'fileroute==X.Y.Z' to select a fixed PyPI version, or uv add ./path/to/fileroute when developing against a local checkout. See uv's project command guide and dependency guide.

As a standalone tool: uvx (an alias for uv tool run) runs Fileroute in its own cached, disposable environment, separate from any project environment. It reads descriptor paths and files from your current working directory, but it does not add Fileroute to the project's dependencies or make it importable by that project's Python code. Use this for ad hoc CLI operations:

uvx fileroute list config/fileroute.yaml
uvx fileroute diagram config/fileroute.yaml
uvx fileroute pull config/fileroute.yaml --dry-run

For a fixed tool version, use uvx --from 'fileroute==X.Y.Z' fileroute --help. See uv's tool guide.

To develop this repository itself:

uv sync
uv run fileroute --help

Configure credentials using .env-sample. SharePoint uses the AZURE_* and SHAREPOINT_* settings; Google supports ADC, service accounts, and user OAuth; S3 uses the standard AWS credential chain. Descriptor loading, diagramming, and dry runs do not authenticate.

fileroute auth login gdrive
fileroute auth login microsoft --auth-mode delegated
fileroute auth login sharepoint --auth-mode delegated

See descriptor diagrams, Google authentication, and the generated CLI and Python API references.

Descriptor model

Four models define the public descriptor API: Catalog, Resource, Location, and CatalogReference. A catalog groups resources and nested catalogs. A resource describes one artifact. Sources and targets are lists of locations.

Field	Meaning
path	Artifact location; a local file/directory for transfers
sources	Upstream inputs or provenance, including local authoring files
targets	Downstream publication destinations
serviceType	Optional location provider: GoogleDrive, SharePoint, or S3
serviceId	Optional provider-native identifier on a location
$ref	Another local catalog document, resolved beside its containing document

For example, a rendered Word document can identify its Quarto source without making that source a publication destination:

$schema: fileroute-catalog
catalogs:
  - name: documentation
    path: docs/_output
    targets:
      - path: https://contoso.sharepoint.com/sites/dev/Shared%20Documents/Docs
        serviceType: SharePoint
    resources:
      - name: guide
        path: docs/_output/guide.docx
        sources:
          - path: docs/guide.qmd
      - name: internal
        path: docs/_output/internal.docx
        targets: []

A retrieval descriptor explicitly identifies its remote source:

resources:
  - name: source-export
    path: downloads/source.csv
    sources:
      - path: s3://my-bucket/exports/source.csv

All artifact paths are relative to the working directory, even when the descriptor is in config/ or references another descriptor. The Python planners also accept an explicit root. Paths are preserved when loading and saving. Transfers reject paths outside that root and symbolic links.

Reference paths are the exception: $ref is relative to the containing descriptor's directory. References stay references on normal load/save. Transfer planning explicitly expands them once; they cannot escape the containing directory. Cycles raise an error.

catalogs:
  - name: research
    $ref: catalogs/research.yaml

Unrecognized metadata fields survive model round trips. Existing YAML descriptors are round-trip edited where practical, retaining comments and authored styles. This is a Fileroute format inspired by Data Package and DCAT, not a full implementation of either standard. $schema is an optional profile label; loading does not fetch a schema from the network.

Common use cases

Each resource has one local artifact path. Its sources record where the artifact came from; its targets list every intended publication destination. One resource can have multiple targets. pull downloads one remote source to path, and push uploads the file at path to all targets. Run them as separate steps; Fileroute does not stream directly between cloud providers or convert source formats. Each example is saved under examples/use-cases/; copy one to config/fileroute.yaml to adapt it to your project.

One SharePoint source and two SharePoint targets

Retrieve a report to the project, then publish copies to two SharePoint sites:

resources:
  - name: monthly-report
    path: artifacts/monthly-report.csv
    sources:
      - path: https://contoso.sharepoint.com/sites/data/Shared%20Documents/monthly-report.csv
    targets:
      - path: https://contoso.sharepoint.com/sites/reports/Shared%20Documents/monthly-report.csv
      - path: https://contoso.sharepoint.com/sites/archive/Shared%20Documents/monthly-report.csv

Generate the Mermaid source from the saved example:

uvx fileroute diagram examples/use-cases/sharepoint-two-targets.yaml --output examples/use-cases/sharepoint-two-targets.mmd

flowchart LR
    n0["artifacts/monthly-report.csv"]
    n1["SharePoint: data/monthly-report.csv"]
    n2["SharePoint: reports/monthly-report.csv"]
    n3["SharePoint: archive/monthly-report.csv"]
    n1 --> n0
    n0 --> n2
    n0 --> n3

For this supported combination, run fileroute pull config/fileroute.yaml and then fileroute push config/fileroute.yaml, with --dry-run on either command to inspect its plan first. Each SharePoint site needs working access.

SharePoint source and Google Drive target

Describe retrieval from SharePoint and intended publication to Google Drive:

resources:
  - name: partner-report
    path: artifacts/partner-report.csv
    sources:
      - path: https://contoso.sharepoint.com/sites/data/Shared%20Documents/partner-report.csv
    targets:
      - path: https://drive.google.com/file/d/GOOGLE_FILE_ID/view

Generate the Mermaid source from the saved example:

uvx fileroute diagram examples/use-cases/sharepoint-google-drive.yaml --output examples/use-cases/sharepoint-google-drive.mmd

flowchart LR
    n0["artifacts/partner-report.csv"]
    n1["SharePoint: data/partner-report.csv"]
    n2["Google Drive: GOOGLE_FILE_ID"]
    n1 --> n0
    n0 --> n2

SharePoint source, SharePoint and Google Drive targets

Use the same local copy for a supported SharePoint publication and a planned Google Drive publication:

resources:
  - name: partner-report
    path: artifacts/partner-report.csv
    sources:
      - path: https://contoso.sharepoint.com/sites/data/Shared%20Documents/partner-report.csv
    targets:
      - path: https://contoso.sharepoint.com/sites/reports/Shared%20Documents/partner-report.csv
      - path: https://drive.google.com/file/d/GOOGLE_FILE_ID/view

Generate the Mermaid source from the saved example:

uvx fileroute diagram examples/use-cases/sharepoint-mixed-targets.yaml --output examples/use-cases/sharepoint-mixed-targets.mmd

flowchart LR
    n0["artifacts/partner-report.csv"]
    n1["SharePoint: data/partner-report.csv"]
    n2["SharePoint: reports/partner-report.csv"]
    n3["Google Drive: GOOGLE_FILE_ID"]
    n1 --> n0
    n0 --> n2
    n0 --> n3

Local authoring source with SharePoint, Google Drive, and S3 targets

Keep the input document as provenance while publishing its rendered output:

resources:
  - name: guide
    path: docs/_output/guide.docx
    sources:
      - path: docs/guide.qmd
    targets:
      - path: https://contoso.sharepoint.com/sites/docs/Shared%20Documents/guide.docx
      - path: https://drive.google.com/file/d/GOOGLE_FILE_ID/view
      - path: s3://example-docs/guide.docx

Generate the Mermaid source from the saved example:

uvx fileroute diagram examples/use-cases/local-three-targets.yaml --output examples/use-cases/local-three-targets.mmd

flowchart LR
    n0["docs/_output/guide.docx"]
    n1["docs/guide.qmd"]
    n2["SharePoint: docs/guide.docx"]
    n3["Google Drive: GOOGLE_FILE_ID"]
    n4["S3: example-docs/guide.docx"]
    n1 --> n0
    n0 --> n2
    n0 --> n3
    n0 --> n4

Render docs/guide.qmd to docs/_output/guide.docx with Quarto before publishing. A local source is provenance; pull does not render or copy it.

Current transfer support: SharePoint, Google Drive, and S3 remote sources can be pulled; only SharePoint targets can be pushed. A descriptor with a Google Drive or S3 target is valid metadata, but push and push --dry-run fail at planning until upload support is implemented. If you need the SharePoint destination now, put it in a separate descriptor (or remove the unsupported targets) for that run. resolve and diagram can still resolve those providers and inspect the descriptor without authentication. Source and target URLs above are examples; replace them with your own accessible files and sites.

Visualize file workflows

diagram renders descriptor intent rather than executing a transfer. It expands local $ref catalogs, resolves known provider URLs in memory, applies catalog target inheritance, and draws sources → artifacts → targets as a standalone SVG. It does not require Graphviz or another rendering dependency.

fileroute diagram config/fileroute.yaml
fileroute diagram config/fileroute.yaml --output docs/fileroute-workflow.svg
fileroute diagram config/fileroute.yaml --output docs/fileroute-workflow.mmd
fileroute diagram config/fileroute.yaml --output docs/fileroute-workflow.html
fileroute diagram config/fileroute.yaml --output docs/fileroute-workflow.md --detail full

Mermaid (.mmd) writes editable flowchart LR source for a fenced mermaid block in GitHub Markdown. The README embeds generated blocks directly; run uv run python scripts/update_readme_diagrams.py after changing an example descriptor. HTML provides a searchable offline file dictionary linked to diagram nodes; Markdown writes an anchored dictionary and companion SVG. --detail summary exports curated metadata; --detail full includes custom fields and service IDs. Inherited targets identify their declaring catalog. Reports link to HTTP(S) locations but do not preview file contents or authenticate. See the diagram guide for anchor stability and duplicate-identity behavior.

The descriptor argument follows the same saved/default selection behavior as pull, push, and resolve, so after fileroute activate you can simply run fileroute diagram. The default output is fileroute-diagram.svg in the current working directory. See Descriptor diagrams for the Python DescriptorGraph API and repository-integration guidance.

Commands

fileroute add export --path downloads/source.csv --source s3://my-bucket/source.csv
fileroute add documentation --catalog --path docs/_output --target https://contoso.sharepoint.com/sites/dev/Docs
fileroute list config/fileroute.yaml --format json
fileroute activate config/fileroute.yaml
fileroute update --name documentation --title "Published documentation"
fileroute resolve config/fileroute.yaml
fileroute resolve config/fileroute.yaml --write
fileroute diagram config/fileroute.yaml
fileroute migrate old.yaml new.yaml --direction push
fileroute pull config/fileroute.yaml --dry-run
fileroute push config/fileroute.yaml --dry-run
fileroute push config/fileroute.yaml

activate saves the active descriptor in .fileroute/descriptor. The optional descriptor argument also accepts an explicit override. update selects an exact name or dot-path and reports ambiguous names; clone descriptor copies a single authored document. For a resource with one source, update --name export --service-type S3 edits that source's provider. Edit sources/targets in YAML for more involved changes, or supply a JSON array using --sources/--targets.

Resolve URLs

resolve previews canonical JSON without authenticating or contacting a remote service. resolve --write saves the result to the selected YAML/JSON descriptor. It infers serviceType from s3:// URLs, drive.google.com / docs.google.com, and SharePoint hosts, while preserving the original URL exactly as a clickable link. For example:

path: https://contoso.sharepoint.com/sites/dev/Docs/guide.docx
serviceType: SharePoint

A conflicting explicit provider is an error. Unrecognized target URLs require an explicit serviceType; local files and general web citations remain valid provider-less sources. Pull requires a supported remote source. Resolution does not follow redirects, fetch remote IDs, or check remote permissions.

Write-back updates only the selected document and preserves $ref entries; resolve referenced descriptors separately to persist their inferred metadata. Repeated resolution is idempotent. Existing YAML comments and styles are retained where practical; byte-for-byte whitespace preservation is not guaranteed. Transfers expand references and resolve the relevant sources or targets in memory, so write-back is optional. push --dry-run and pull --dry-run provide concrete transfer plans; there is no separate plan command or stored lockfile.

Push

push publishes path to targets. Sources are never treated as targets. Only SharePoint uploads are currently implemented. Other providers fail during planning, before authentication.

• Catalog targets denote existing remote folders. Children inherit those targets using paths relative to the declaring catalog's path (or the working root when no catalog path is given).
• A child's explicit targets replace inherited targets. targets: [] disables publication for that child and its descendants unless a descendant declares its own targets.
• A catalog with children publishes only its declared children. A leaf catalog with a path publishes all files in that directory tree.
• Explicit resource targets are exact file URLs, allowing renaming. Set a target's entityType: Directory to append the local filename to a folder URL.
• Multiple targets publish multiple copies. Conflicting files aimed at the same destination fail before any transfer.

The planner checks all inputs before authentication. SharePoint transfers create missing child folders and create/replace files; they never delete remote files. Files above Microsoft Graph's 250 MB single-request limit fail planning. Dry runs show the proposed destinations without contacting the remote service; they cannot verify remote permissions or folder existence.

Pull

pull materializes one remote sources entry into each artifact's path. A leaf catalog can retrieve a whole remote directory. A catalog with children retrieves only those children. Targets never affect retrieval.

Multiple sources may represent a transformation, so pull refuses to choose one. Local provenance such as .qmd inputs is not a download operation: build those artifacts with their authoring tool, then push them. Use separate catalogs for retrieval and publication when their source semantics differ. Dry runs plan remote directories as units; execution enumerates and checks remote file paths before writing any files. Pull replaces existing local files at planned paths.

Python API and architecture

from pathlib import Path
from fileroute import Catalog, Resource, Location
from fileroute.descriptor import save
from fileroute.diagram import load_graph, render_svg
from fileroute.transfer import plan_push, push

catalog = Catalog(resources=[Resource(
    name="guide", path="docs/guide.docx",
    sources=[Location(path="docs/guide.qmd")],
    targets=[Location(path="https://contoso.sharepoint.com/sites/dev/Docs/guide.docx")],
)])
save(catalog, "config/fileroute.yaml")
graph = load_graph("config/fileroute.yaml")
render_svg(graph, "fileroute-diagram.svg")
plan = plan_push(Path("config/fileroute.yaml"), root=Path.cwd())
# Inspect plan before transfer.
push(plan)

The core has one module per responsibility:

• models.py: declarative Catalog, Resource, Location, CatalogReference and ServiceType validation.
• descriptor.py: load, save, walk, find, and offline resolve.
• diagram.py: semantic descriptor graphs and dependency-free SVG rendering.
• diagram_reports.py: offline HTML inspector and Markdown file dictionary.
• migration.py: explicit one-way conversion of legacy descriptors.
• transfer.py: plan_pull / plan_push, then pull / push execution.
• item.py: runtime ServiceItem hierarchy.
• clients/sharepoint.py, clients/googledrive.py, clients/s3.py: provider APIs.
• auth/ and commands/: credential handling and CLI workflows.

After resolution, location.service_type is the authoritative ServiceType enum used directly for provider dispatch. Pydantic accepts aliases such as gdrive or google_drive; saved values are GoogleDrive, SharePoint, or S3. Python attributes use service_type, service_id, and entity_type; descriptors use serviceType, serviceId, and entityType. No provider-string conversion layer or dynamic registry is needed.

ServiceItem and provider clients retain their runtime roles. Clients own provider-specific authentication and HTTP behavior; runtime items expose refresh(), children, get_path(), iter_items(), iter_files(), and download(). item.to_catalog() emits canonical artifact paths and sources. Runtime paths remain relative to the provider container. get_path() is relative to the current item. Parent relationships and recursive traversal use a snapshot until refresh or mutation invalidates it.

Breaking API changes

Use the four models above and the functions in descriptor, diagram, and transfer. Model I/O and traversal methods, upload.py, download.py, helpers.py, and the provider registry have been removed. The CLI has no upload alias or set command. Run activate again to select a descriptor using the new single-path selection file; obsolete saved workflow defaults are no longer read.

Legacy Drive* classes, packages, _cache, and artifact-level provider fields are unsupported. Update authored descriptors to path, sources, targets, resources, and catalogs; provider metadata belongs on a location. Normal loading has no compatibility shims or automatic legacy conversion. Use fileroute migrate OLD NEW --direction pull|push for an explicit, one-way conversion that preserves the input file.

Development

poe check
poe docs-check
poe docs-update
poe docs-build
poe docs-serve

Install the task runner with uv tool install poethepoet==0.48.0, or use uvx --from poethepoet==0.48.0 poe <task> for a one-off run. poe docs-build installs the documentation extra through uv run.

Provider tests use mocks; they do not prove live tenant permissions or transfers.

Field naming references: Data Resource, DCAT, and OpenMetadata Drive Service.

Package releases

Install the repository task runner with uv tool install poethepoet==0.48.0, then run poe release-check for validation and poe build for a local wheel and source distribution. One-off usage is uvx --from poethepoet==0.48.0 poe release-check.

The publish-to-pypi.yaml workflow has three jobs: prepare, publish, and release. Pushes to dev start or increment a patch development version (1.0.0 → 1.0.1.dev1 → 1.0.1.dev2). Pushes to main promote a prerelease to stable, or increment the patch when the source already has a stable version. After a stable tag exists, dev starts the next patch series. Normal releases should not edit the version manually. Merge updated main back into dev when needed to keep the branches' version baselines aligned.

poe release --branch dev --source <full-commit-sha> runs scripts/release.py in a clean checkout of that source. It uses uv version --no-sync to update pyproject.toml and uv.lock, invokes poe build, and creates a local release commit and annotated tag. This is the CI preparation command: it changes the local checkout but does not push or publish. The workflow first saves the built distributions, then atomically pushes the release commit and tag. Both branches share one release concurrency group; a stale run fails rather than overwriting newer work. Rapid pushes may supersede pending runs; the latest source should be released. Workflow pushes use GITHUB_TOKEN and do not recursively trigger another push workflow.

PyPI publication uses Trusted Publishing bound to this repository and publish-to-pypi.yaml. No long-lived PyPI token is needed, and the publish job does not require a GitHub environment unless the PyPI publisher is configured to expect one. Only stable main versions get a GitHub Release, after PyPI succeeds. The branch rules must permit the workflow's version commit; rejected pushes leave both remote refs unchanged.

For a failed publish, use Re-run failed jobs on the original Actions run. The publish job downloads the saved wheel and source distribution without rebuilding. Identical PyPI uploads can be retried, including a partially completed upload. A full rerun recognizes the tagged source/branch before calculating a version and finds its unexpired artifact in the original workflow run by source commit. Existing GitHub Releases are left intact. If preparation failed before the atomic push, a fresh attempt can rebuild and replace that run's unpublished artifact. If the original artifact has expired or been deleted after the push, stop and recover those exact files; the workflow deliberately does not rebuild a published version.

poe release-test exercises the release helper with temporary local Git remotes and real uv version changes; distribution builds are mocked. No test publishes packages or contacts cloud providers.