README.md

metadata

title: Cohere Parse
sdk: gradio
sdk_version: 6.21.0
python_version: 3.12
app_file: app.py
pinned: false

Cohere Parse

An account-free document parsing playground for Cohere Parse. Upload a PDF or image, inspect the normalized page render, and review extracted text, tables, and bounding boxes.

Behavior

• Accepts one PDF, PNG, JPEG, or WebP up to 20 MB.
• Renders documents at up to 200 DPI and a 2,048 pixel maximum edge.
• Processes the first 10 pages and warns when additional pages are omitted.
• Re-encodes every page as lossless PNG before calling the Cohere API.
• Sends image-only /v2/parse requests with structured block output and does not persist parser output.

Configuration

Set these in the Hugging Face Space's Settings page. Store API keys as Space secrets; the other values can be public Space variables.

Name	Default	Type
COHERE_ENV	staging	Variable
COHERE_API_KEY	required in production	Secret
COHERE_STG_API_KEY	required in staging	Secret
COHERE_API_URL	selected from COHERE_ENV	Optional variable override
COHERE_MODEL	parse-v5.0	Variable
COHERE_TIMEOUT	300	Variable

COHERE_ENV=production uses COHERE_API_KEY and https://api.cohere.com/v2/parse. COHERE_ENV=staging uses COHERE_STG_API_KEY and https://stg.api.cohere.ai/v2/parse. Keep COHERE_API_URL unset unless routing through an intentional custom endpoint.

Use a dedicated, revocable API key with a quota appropriate for a public demo.

Local development

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt -r requirements-dev.txt
cp .env.example .env
set -a && source .env && set +a
python app.py

Open http://127.0.0.1:7860.

Checks

pytest
python -m compileall app.py document_processing.py parser_client.py parser_output.py rate_limit.py ui_helpers.py

An integration test against staging is skipped unless RUN_STAGING_INTEGRATION=1, COHERE_STG_API_KEY is configured, and a local sample document is available.

Deploy

Create a private Gradio Space under CohereLabs, configure the secret and variables above, then upload this repository:

hf auth login
hf upload CohereLabs/cohere-parse . . --repo-type space

Every commit uploaded to the Space rebuilds and restarts the app. Keep the Space private during review, then switch it to protected visibility for a public app with private source.

Privacy and security

Uploaded files are decoded in the Space, converted to normalized PNG page images, and deleted after preparation. Only those page images are sent to the Cohere API. Gradio may temporarily cache uploaded files while processing them; the app deletes source files explicitly and also configures short cache cleanup.

The application also enforces upload, page, pixel, queue, concurrency, and per-client request limits. These controls reduce abuse but do not replace a gateway-enforced quota on the API credential.