unstract-cli

unstract runs LLMWhisperer text extraction and Unstract API deployments from the terminal. Pass -o json and every command prints one JSON envelope, so a shell script or a coding agent can drive it.

Full reference: https://docs.unstract.com/unstract/unstract_platform/cli/unstract_cli/

curl -LsSf https://raw.githubusercontent.com/Zipstack/unstract-cli/main/install.sh | sh

The installer fetches uv if it is missing and installs the CLI with it; uv brings its own Python. With uv or pip already there:

uv tool install unstract-cli   # or: pip install unstract-cli
unstract --version

unstract auth login

A wizard asks for each product's URL (Enter keeps the cloud host; self-hosted, type your own) and API keys, then writes ~/.unstract/config.toml. Then verify:

unstract auth whoami             # which organisation the platform key belongs to
unstract config doctor --probe   # where each setting resolved from, keys checked

# Extract text from a document (path or URL); waits for the result
unstract whisper extract invoice.pdf -o raw > invoice.txt

# What deployments can I run?
unstract docstudio deployment ls

# Run one and wait for the structured result
unstract docstudio deployment run invoice-parser invoice.pdf

# Long job: submit, then check later
unstract docstudio deployment run invoice-parser invoice.pdf --no-wait
unstract docstudio deployment status invoice-parser <execution_id>

--help on any command lists its options; unstract --discover full prints the whole command tree, every flag and the exit-code table as JSON.

~/.unstract/config.toml, or $UNSTRACT_CONFIG, or --config, or a project-local .unstract.toml is found by upward search. Here's an example config that uses environment variables for the API keys.

default_profile = "cloud-us"

[profiles.cloud-us.docstudio]
base_url = "https://us-central.unstract.com"
org_id = "org_ABC123"
platform_key = "env:UNSTRACT_PLATFORM_KEY"
api_key = "env:UNSTRACT_DEPLOYMENT_KEY"

[profiles.cloud-us.llmwhisperer]
base_url = "https://llmwhisperer-api.us-central.unstract.com/api/v2"
api_key = "env:LLMWHISPERER_API_KEY"

# Only for a deployment whose key differs from the profile's.
[profiles.cloud-us.deployments."invoice-parser"]
api_key = "env:INVOICE_PARSER_KEY"

Every setting resolves flag > environment > profile > built-in default. auth login writes keys literally; env:VAR_NAME keeps them out of the file. The connection flags on each group (--base-url, --api-key, --org-id, --platform-key) override the profile for one invocation without writing anything. config doctor reports where each setting resolved from without echoing a value, and exits non-zero when one of its checks fails.

The same settings without a file, for CI, containers and agents:

export UNSTRACT_PLATFORM_KEY=...      # auth whoami, deployment ls
export UNSTRACT_DEPLOYMENT_KEY=...    # deployment run / status
export UNSTRACT_ORG_ID=...            # the organisation id auth whoami reports
export LLMWHISPERER_API_KEY=...       # whisper …
export UNSTRACT_BASE_URL=...          # self-hosted only
export LLMWHISPERER_BASE_URL=...      # self-hosted only

auth login also takes each key as a flag (--platform-key, --deployment-key, --llmwhisperer-key; - reads it from stdin) and the host as --base-url, so it runs without a terminal too.

Parsing anything? Pass -o json. stdout then carries exactly one envelope, on success and on failure alike, and diagnostics go to stderr:

{"ok": true, "data": {...}, "error": null, "meta": {"contract_version": 1}}

Ignore fields you do not recognise; refuse a meta.contract_version above the one you were written against. -o raw prints one field unwrapped. When a coding agent is driving (detected from the environment it sets) json is the default; --agent yes|no forces that, and an explicit -o wins over both.

Failures exit non-zero with a stable code:

uv sync --extra dev
uv run pytest
uv run ruff check .

On Slack, join great conversations around LLMs, their ecosystem and leveraging them to automate the previously unautomatable!

Unstract Cloud: Signup and Try!