binctl is a tiny graph-based inventory system for people with too many bins, boxes, and shelves.
Instead of thinking in terms of "SKUs" and "stock levels", binctl models your world as:
- nodes - items, bins, shelves, rooms, etc.
- edges - "this thing lives inside that thing".
It's backed by a Flask API with a CLI frontend.
pip install binctl-serverInstall database drivers only when needed:
- MySQL:
pip install 'binctl-server[mysql]' - PostgreSQL:
pip install 'binctl-server[postgresql]'
The fastest path to a running system. Requires Python 3.11+. Run each block in your terminal:
# 1. Generate the ignored API client
./genclient.sh
# 2. Create and activate a virtual environment
python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
# 3. Install dependencies — binctl-client is installed from PyPI
pip install -e .
# 4. Configure the database — copy .env.example and uncomment the SQLite line
cp .env.example .env
# In .env, uncomment: DATABASE_URL=sqlite:///binctl.db
# 5. Initialize the database
binctl-manage init-db
# 6. Create a user and get a token
binctl-manage create-user alice --token
# → prints something like: token: abc123...
export TOKEN=<paste token here>
# 7. Start the server (keep this terminal open, or run it in the background)
uvicorn binctl_server.web:create_app --factory
# 8. In another terminal (with .venv activated), verify it works
binctl --token $TOKEN node list-
Generate the API client, then create and activate a virtual environment and install dependencies:
./genclient.sh python3 -m venv .venv source .venv/bin/activate # Windows: .venv\Scripts\activate pip install -e .binctl-clientis installed from PyPI.genclient.shusesuvxto generate a local client frombinctl_server/openapi.yamlas an API-contract check; itsbinctl-client/output is ignored and must not be committed.If you need MySQL or PostgreSQL support, install the optional driver afterward:
- MySQL:
pip install pymysqlorpip install 'binctl-server[mysql]' - PostgreSQL:
pip install psycopg2-binaryorpip install 'binctl-server[postgresql]' - SQLite — no extra driver needed, skip this step.
- MySQL:
-
Copy
.env.exampleand set your database URL:cp .env.example .envSupported URL formats:
- SQLite:
sqlite:///binctl.db - MySQL:
mysql+pymysql://user:password@localhost/binctl - PostgreSQL:
postgresql+psycopg2://user:password@localhost/binctl
The server and
binctl-manageload.envautomatically — no manualexportneeded. - SQLite:
-
Initialize the database:
binctl-manage init-db -
Start the server:
uvicorn binctl_server.web:create_app --factory -
Create a user:
- With a non-expiring API token (recommended for scripts):
binctl-manage create-user alice --token - With a password (for browser/interactive use):
binctl-manage create-user alice --password <password>
Then pass
--token <token>(or--username/--password) tobinctlcommands. - With a non-expiring API token (recommended for scripts):
binctl node list|get|create|update|delete- manage nodesbinctl tag list|get|create|update|delete- manage tags by unique name
Key flags for binctl:
| Flag | Description |
|---|---|
--base-url |
API server URL (default: http://localhost:5000) |
--token |
Bearer token (preferred) |
--username / --password |
Login-based auth |
binctl-manage subcommands (server-side user/token management):
binctl-manage init-db- initialize the database schemabinctl-manage create-user <username> --password <p>- create a user with a passwordbinctl-manage create-user <username> --token- create a passwordless user and emit a non-expiring tokenbinctl-manage set-password <username>- interactively set a new password for a userbinctl-manage list-users- list usersbinctl-manage list-tokens <username>- list tokens for a userbinctl-manage revoke-tokens <username>- revoke all tokens for a user
This walks through creating a hierarchy (room → shelf → item), listing it, and moving a node.
Node IDs are opaque strings (e.g. wGIDjZ0AAC), not sequential integers. Copy them from the
output of each command — do not guess or construct them by hand.
export TOKEN=<your token>
# Create a room (a container)
binctl --token $TOKEN node create --label "Garage" --is-container
# → {"id": "wGIDjZ0AAC", "label": "Garage", "is_container": true, ...}
# Create a shelf inside the room — use the id from the previous output
binctl --token $TOKEN node create --label "Shelf A" --is-container --parent-id wGIDjZ0AAC
# → {"id": "xK3mPq1BBD", "label": "Shelf A", ...}
# Add an item to the shelf
binctl --token $TOKEN node create --label "Power drill" --parent-id xK3mPq1BBD
# → {"id": "yR7nTs2CCE", "label": "Power drill", ...}
# List everything
binctl --token $TOKEN node list
# Move the drill to a different shelf (use that shelf's id from its create output)
binctl --token $TOKEN node update --node-id yR7nTs2CCE --parent-id <other-shelf-id>
# Detach the drill entirely (no parent, becomes a root node)
binctl --token $TOKEN node update --node-id yR7nTs2CCE --no-parentA systemd unit file ships with binctl-server for running the server persistently on a Linux
host: at systemd/binctl-server.service in this repo, or at
<venv>/share/binctl/systemd/binctl-server.service if installed from PyPI. Copy it to
/etc/systemd/system/, adjust the paths/user, then
systemctl daemon-reload && systemctl enable --now binctl-server. See
Server administration for the full walkthrough.
binctl is not designed to be exposed to the internet. It is intended for use on trusted local or private networks. Security bugs will be fixed when found, but the threat model assumes a trusted network environment.
If you must expose the server to untrusted networks, place a reverse proxy such as nginx or Caddy in front of it to handle:
- SSL/TLS termination — the Flask/uvicorn server does not handle TLS on its own.
- Rate limiting on
POST /v1/auth/login— scrypt makes each login attempt slow, but a determined attacker can still brute-force credentials over time without a request rate limit.
Token security — tokens created via the login API (web/browser sessions) expire after 30 days by default; set SESSION_LIFETIME_DAYS to override. Tokens created via binctl-manage create-user do not expire by default. All tokens should be treated with the same secrecy as a password: store them safely, do not share them, and revoke compromised tokens promptly with binctl-manage revoke-tokens.
Password policy — set-password enforces a minimum of 6 characters. No other complexity requirements are enforced by the application. Choose a strong password of at least 16 characters using a mix of uppercase letters, lowercase letters, digits, and symbols.
Contributors should use uv for a faster workflow. Generate the
ignored local client first, then install the published binctl-client package and other
development dependencies.
# Install uv (skip if already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh
# Generate the client from the API contract
./genclient.sh
# Install all dependencies including dev tools
uv sync
# Run checks
uv run pytest
uv run ruff check
uv run ruff format --check
PYTHONPATH=.:binctl-client uv run ty checkThe following must pass before merging (run with the venv activated):
pytest
ruff check
ruff format --check
PYTHONPATH=.:binctl-client ty check
In GitHub Actions, run the Publish to PyPI workflow and select the patch,
minor, or major version increment. The workflow runs from main, uses
uv version --bump to update pyproject.toml and uv.lock, commits the new
version, creates and pushes its matching v<version> tag, and publishes both
binctl-server and binctl-client in the same run. The workflow regenerates
binctl-client, syncs its version to the server version, then builds and
publishes both distributions.
Before the first release, configure PyPI Trusted Publishing for
clueboard/binctl with workflow filename publish.yml and environment name
pypi for both projects (binctl-server and binctl-client). The workflow
uses GitHub Actions OIDC, so it does not require a PyPI API token or
repository secret.
Everything is a node:
- items (tools, parts, one-off widgets)
- containers (bins, boxes, drawers)
- higher-level containers (shelves, rooms, buildings)
Nodes live in the nodes table:
id- BIGINT primary keylabel- human-readable namedescription- optional free textis_container-TRUEif this node can contain children
Containment is modeled via the edges table:
parent_id→child_id- each child has at most one parent (enforced by a UNIQUE constraint)
- self-loops are rejected (
parent_id <> child_id)
This gives you a forest of trees:
- top-level nodes (rooms, "unplaced items") have no parent
- containers and items have exactly one parent
- containers can have many children
When you delete a container, any direct children can be reassigned by setting ORPHAN_LOCATION in your .env. The server looks up a container by that label at delete time, and if none exists outside the deleted subtree it creates a new root container with that label automatically.
Tags are stored in tags + tag_node for future filtering and categorization.
Tags are exposed via binctl tag list|get|create|update.
The server normally assigns node IDs with POST /v1/nodes. Offline clients may generate an ID
and create or fully replace that resource with PUT /v1/nodes/{id}. A PUT body is the complete
desired state: omitted description, parent_id, and tags are cleared, and omitted
is_container is false. Upload parents before children.
Node IDs are canonical, case-sensitive base62 encodings using
0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz. The decoded signed 64-bit integer is
(microseconds_since_2020-03-11T00:00:00Z << 12) | secure_random_12_bits; valid IDs decode to
1...(2^63-1) with no leading zero. Client clock skew is accepted.
Idempotency is controlled by a database-wide lock. PUT /v1/lock enables it, GET /v1/lock
reports its state, and DELETE /v1/lock disables it. While enabled, every node or tag mutation
must include an Idempotency-Key; missing keys return 428 Precondition Required. While disabled,
mutation keys are rejected with 409 Conflict. Repeating the same authenticated request with the
same key returns its original successful response; reusing a key for a different method, path, or
JSON body returns 409 Conflict. Failed requests do not consume the key. The lock endpoints are
exempt from the lock so it can always be disabled.
Tags are public unique strings matching [a-z-]+ rather than public IDs. Node write bodies use
tags: ["fragile"], and missing tags are created atomically with the node mutation.
Clients can initialize their local graph with GET /v1/snapshot. The authenticated endpoint
returns every node, including its parent ID and assigned tag names, in deterministic ID order.
The snapshot is read from one database statement so concurrent writes cannot produce a mixture
of graph states. Unattached tags are not included. The response also includes event_cursor.
To receive subsequent inventory changes, connect to authenticated GET /v1/events with that
cursor in the Last-Event-ID header. The endpoint replays newer durable events and then remains
open as a Server-Sent Events stream. If the cursor is older than retained history, the server
sends reset-required and closes the stream; fetch a new snapshot before reconnecting. The
generated Python package exports synchronous stream_events() and asynchronous
astream_events() helpers for this workflow.
ModuleNotFoundError: No module named 'binctl_client'
Reinstall the project dependencies with pip install -e . or uv sync.
Connection refused / Failed to connect when running binctl commands
The server is not running. Start it with uvicorn binctl_server.web:create_app --factory (binds to
http://localhost:5000 by default). If you changed the port, pass --base-url http://localhost:<port>
to binctl.
401 Unauthorized
Token is missing or wrong. Re-create one with binctl-manage create-user alice --token,
or list existing tokens with binctl-manage list-tokens alice.
DATABASE_URL not set or database errors on startup
Export the variable before running server or manage commands:
export DATABASE_URL=sqlite:///binctl.db
Or add it to your .env file (copied from .env.example).