ContextQMD
Back to Activepieces

Jungle Grid

packages/pieces/community/jungle-grid/README.md

0.85.43.3 KB
Original Source

Jungle Grid

Jungle Grid runs and monitors AI workloads on managed GPU capacity. Use this piece to estimate a workload, submit it for async execution, poll status/runtime/logs, retrieve artifacts, and cancel non-terminal jobs from an Activepieces workflow.

Setup

  1. Create or sign in to a Jungle Grid account at https://junglegrid.dev.
  2. Generate a scoped API key in the Jungle Grid portal.
  3. Add a Jungle Grid connection in Activepieces.
  4. Paste the API key into the API Key field.
  5. Keep API Base URL as https://api.junglegrid.dev unless Jungle Grid gives your workspace a different endpoint.

Recommended API key scopes:

  • jobs:estimate for Estimate Job.
  • jobs:submit or jobs:write for Submit Job and Cancel Job.
  • jobs:read or jobs:write for List Jobs, Get Job Status, Get Job Runtime, logs, and artifacts.
  • logs:read for Get Job Logs when not using jobs:read.

Actions

  • Estimate Job: previews route, capacity, queue, and cost signals without starting compute.
  • Submit Job: queues a workload and returns immediately with job metadata.
  • List Jobs: finds recent jobs and job IDs, optionally filtered by status.
  • Get Job Status: reads the current status and details for a job.
  • Get Job Runtime: reads runtime tails, exit code, and runtime availability details when available.
  • Get Job Logs: fetches recent stdout, stderr, or combined log entries.
  • List Job Artifacts: lists files produced by a job.
  • Get Artifact Download URL: creates a temporary signed URL for one artifact.
  • Cancel Job: cancels a non-terminal job and may stop active execution.

Examples

Estimate before running:

  • Workload Type: batch
  • Container Image: python:3.11
  • Model Size (GB): 1
  • Command: python
  • Arguments: -c, print(42)
  • Optimize For: balanced

Submit an inference workload:

  • Workload Type: inference
  • Container Image: pytorch/pytorch:2.4.0-cuda12.1-cudnn9-runtime
  • Model Size (GB): 7
  • Job Name: chat-infer

Submit a batch workload:

  • Workload Type: batch
  • Container Image: python:3.11
  • Model Size (GB): 1
  • Command: python
  • Arguments: -c, print(42)
  • Optimize For: cost

Monitor a job:

  1. Use Submit Job.
  2. Store job_id from the response.
  3. Add a Delay step.
  4. Use Get Job Status, Get Job Runtime, or Get Job Logs with job_id.
  5. Repeat or branch until the job reaches completed, failed, rejected, or cancelled.

Retrieve logs and artifacts:

  1. Use Get Job Logs with job_id, Tail Lines, and Stream.
  2. After completion, use List Job Artifacts.
  3. Use Get Artifact Download URL with job_id and artifact_id.
  4. Treat the returned signed URL as temporary and sensitive.

Trigger Jungle Grid from an automation:

  1. Start with any Activepieces trigger, such as a webhook, schedule, form submission, or new database row.
  2. Use Estimate Job when the workflow needs a cost or capacity preview.
  3. Use Submit Job to queue work.
  4. Persist job_id in the downstream system or poll with Delay + Get Job Status.
  5. Use runtime, logs, and artifacts actions to collect outputs for notifications, storage, or follow-up processing.