the platform for async ai workflows
Describe your workflows once, then let Exosphere run them in the background with up to 75 % lower cost—built for jobs that must run reliably at scale.
Create Multi Step AI Workflows
Connect tools, models, and APIs to automate complex async jobs.
PyTorch model weights containing the trained parameters of a deep learning model for natural language processing.
Dataset containing labeled examples used for training machine learning models, including features and target variables.
Configuration file defining hyperparameters, model architecture, and training settings for AI model deployment.
Documentation for the API endpoints and their usage, including request parameters, response formats, and examples.
Performance metrics and evaluation results of AI models, including accuracy, precision, recall, and F1 scores.
PyTorch model weights containing the trained parameters of a deep learning model for natural language processing.
Dataset containing labeled examples used for training machine learning models, including features and target variables.
Configuration file defining hyperparameters, model architecture, and training settings for AI model deployment.
Documentation for the API endpoints and their usage, including request parameters, response formats, and examples.
Performance metrics and evaluation results of AI models, including accuracy, precision, recall, and F1 scores.
PyTorch model weights containing the trained parameters of a deep learning model for natural language processing.
Dataset containing labeled examples used for training machine learning models, including features and target variables.
Configuration file defining hyperparameters, model architecture, and training settings for AI model deployment.
Documentation for the API endpoints and their usage, including request parameters, response formats, and examples.
Performance metrics and evaluation results of AI models, including accuracy, precision, recall, and F1 scores.
PyTorch model weights containing the trained parameters of a deep learning model for natural language processing.
Dataset containing labeled examples used for training machine learning models, including features and target variables.
Configuration file defining hyperparameters, model architecture, and training settings for AI model deployment.
Documentation for the API endpoints and their usage, including request parameters, response formats, and examples.
Performance metrics and evaluation results of AI models, including accuracy, precision, recall, and F1 scores.
Work with your own files
Upload, process, and analyze files directly in your workflows.From PDFs to CSVs, bring your own data.
💸
6 hour SLA
Define SLAs for your jobs
Set job deadlines. Exosphere manages batching, retries, and cost optimization.
orbit sdk
open-source sdk for async ai jobs
Define Python functions. Orbit handles retries, scaling, and orchestration.
Higher SLAs = Lower Costs
Trade off latency for price. Define SLAs and let Exosphere optimize cost.
get running within two lines of code
Say goodbye to managing your own code to orchestrate ai workflows.
workflow.py
import requests
def load_files_from_s3(bucket, keys):
print("Loading files from S3...")
res = requests.post("https://api.example.com/s3/generate-urls", json={
"bucket": bucket,
"keys": keys
})
res.raise_for_status()
return res.json()["urls"]
def send_inference_request(deployment_id, file_urls):
print("Sending inference request to Gemma...")
res = requests.post("https://api.example.com/gemma/infer", json={
"deployment_id": deployment_id,
"files": file_urls
})
res.raise_for_status()
return res.json()["job_id"]
def poll_for_completion(job_id, interval=5, max_retries=20):
print("Polling for job completion...")
for attempt in range(max_retries):
res = requests.get(f"https://api.example.com/gemma/status/{job_id}")
res.raise_for_status()
status = res.json()["status"]
if status == "completed":
return True
elif status == "failed":
raise Exception("Job failed")
time.sleep(interval)
raise TimeoutError("Polling timed out")
def fetch_inference_result(job_id):
print("Fetching inference result...")
res = requests.get(f"https://api.example.com/gemma/result/{job_id}")
res.raise_for_status()
return res.json()
def main(string bucket, string[] keys):
file_urls = with_retries(load_files_from_s3, bucket, keys) # [!code highlight]
job_id = with_retries(send_inference_request, deployment_id, file_urls) # [!code highlight]
if with_retries(poll_for_completion, job_id):
result = with_retries(fetch_inference_result, job_id) # [!code highlight]
ocr_result = with_retries(run_ocr_on_images, result["images"]) # [!code highlight]
summary = summarize_output(ocr_result) # [!code highlight]
return summary
if __name__ == "__main__":
main()workflow.py
from exosphere import run_workflow
def main():
result = run_workflow( # [!code focus]
"gemma_gpu_pipeline.yaml", # [!code focus]
input={
"bucket": "user-bucket", # [!code focus]
"files": ["doc1.pdf", "doc2.pdf"] # [!code focus]
},
sla_minutes=360 # [!code focus]
)
print("Final Summary:", result)
if __name__ == "__main__":
main()
batch and save on agents
Save up to 60% thanks to batching optimizations
reliability first, always
For workloads that need to be running
launch your ai workflows
Book a call with our team to discuss how we can help you build and optimize your AI workflows for maximum efficiency.