LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Python ai-agentsdata-engineeringdocument-aigenerative-aiidpjson-extractionllmmcp-serverocrpdf-extractionprompt-engineeringstructured-output
Unstract Runner
This module contains the logic for how Tools are executed in DND mode.
Docker build and publish
Build docker image
docker build -t unstract/runner:1.1.1 .
Then publish it into Docker hub
docker push unstract/runner:1.1.1
Run docker compose
To expose the Unstract runner service.
docker compose -f docker-compose.yaml up
Using with Docker Desktop
The runner makes use of the python docker client's from_env() to initialize the docker client. If you are using Docker Desktop, this might not work as expected due to the issue described here. In that case follow the resolutions from the issue
- Either set
DOCKER_HOSTenv to wherever the docker endpoint is configured.
export DOCKER_HOST=unix:///$HOME/.docker/desktop/docker.sock
- Or create the symlink manually (as recommended in the issue)
sudo ln -s "$HOME/.docker/run/docker.sock" /var/run/docker.sock
Required Environment Variables
| Variable | Description |
|---|---|
CELERY_BROKER_BASE_URL |
Base URL for Celery's message broker, used to queue tasks. Must match backend configuration. |
CELERY_BROKER_USER |
Username for Celery's message broker. |
CELERY_BROKER_PASS |
Password for Celery's message broker. |
TOOL_CONTAINER_NETWORK |
Network used for running tool containers. |
TOOL_CONTAINER_LABELS |
Labels applied to tool containers for observability [Optional]. |
EXECUTION_DATA_DIR |
Target mount directory within tool containers. (Default: "/data") |
LOG_LEVEL |
Log level for runner (Options: INFO, WARNING, ERROR, DEBUG, etc.) |
REMOVE_CONTAINER_ON_EXIT |
Flag to decide whether to clean up/ remove the tool container after execution. (Default: True) |