Symbioza
Menu

CLOUD GPU JOBS FOR AI AGENTS

The run your laptop
won’t handle.

Your AI agent sends training, evaluation and batch inference to a cloud GPU without leaving the conversation.
Symbioza manages the compute. You set the spending limit.

Free estimate with an account. Runs use prepaid credit — from $10, no subscription.

YOUR CODE · YOUR DATA · YOUR SPENDING LIMITExplore what happens next ↓

01 / THE HANDOFF

A request.
A result. A receipt.

Describe the job, review an estimate, and check the files and charges when it ends.

EXAMPLE / MODEL EVALUATION
“Run my evaluation.
Keep it under $2.”

You set the work and the limit. Your agent carries the request through Symbioza.

YOUR $2 SPENDING LIMIT
evaluation.pycode + data + expected files
YOUR JOB. YOUR SPENDING LIMIT.
Illustrative journey · sample figures
From request to resultILLUSTRATIVE
Explore the sample run
YOU → YOUR AGENT
Run this model evaluation. Return the metrics and don’t spend more than $2.

You provide the code and data. Your agent prepares the job and asks for an estimate.

review before spending

An estimate before you submit.

Estimated cost$0.84
Spending limit$2.00

An estimate books no compute and is not a final price. Your spending limit covers the total job charge, including retries.

delivered

The files. And the charge.

Final charge · this example$0.72
Files delivered2 / 2
metrics.json
predictions.csv

File delivery and billing have separate statuses. A charge can remain provisional after a job ends.

Sample journey, not a live job or price quote. Select a step above.

CONNECT YOUR ASSISTANT

Your workflow.
More compute.

Run GPU jobs from the assistant you already use. Get an estimate, set your spending limit and collect the results through Symbioza.

Free estimates · Manual MCP setup for other clients

A spending limit you control
No machine to manage
Files with verifiable hashes

02 / WORK TO RESULTS

Big ideas.
Heavy lifting, handled.

Bring your container image, the command to run and where the data is. Define the output files you want to collect.

YOUR WORK
TRAININGEVALUATIONBATCH INFERENCE
01 / DESCRIBE & ESTIMATE

Your code.
Your data.
Your limit.

Ask your agent for a free estimate. Review it before submitting a run that spends prepaid credit.

02 / RUN & DISCONNECT

We handle
the machine.

Symbioza manages the run. Recovery follows your job policy and remaining budget.

03 / COLLECT & CHECK

Work, completed.

weights + logsmetrics + predictionsprocessed data

EXAMPLE OUTPUTS + RUN RECEIPT · FILES AND CHARGES REPORTED SEPARATELY

Fine-tune a model

Run your training script on cloud compute. Save model weights and logs as outputs.

See a sample request

“Run my training script on this dataset. Save the model and metrics. Keep the whole run under $10.”

Bring a container and training code. Checkpoint recovery requires compatible save-and-resume logic.

Evaluate a model

Run your evaluation suite and get the numbers back in a repeatable format.

See a sample request

“Run my evaluation harness and return metrics.json. Estimate first, with a $2 spending limit.”

Use the same model, dataset and evaluation code when comparing runs. Check the run record for execution details.

Process a dataset

Run batch inference using a dataset URL. Collect predictions as output files.

See a sample request

“Score this dataset using my container. Read it from this URL and return predictions.csv.”

Declare the data and output paths. The estimate depends on the actual workload and transfer needs.

03 / YOUR FIRST ESTIMATE

Know the cost.
Then make the call.

Start with a free GPU job estimate. Review the expected cost and workload requirements with your agent before you add credit for a run.

YOUR JOBYOUR ESTIMATE

An estimate books no compute.

YOUR AGENT + SYMBIOZAFREE ESTIMATE
01

Connect your agent

Use Claude Code or a compatible MCP client. Add the connector, then follow the client’s sign-in flow.

Using Claude Code?

Claude Code: install the plugin — claude plugin marketplace add symbioza/claude-plugin then claude plugin install symbioza@symbioza. Other MCP clients: add the connector URL.

MCP connects your agent to Symbioza’s tools. Your client must support remote HTTP connections and sign-in.

The Claude Code plugin, ChatGPT and other clients: Integrations →

02

Bring a job. Ask for a price.

ESTIMATE FIRST

Have your container, command, data location and output paths ready. Your agent can help prepare the request.

Copy the URL or example prompt to use in your agent.

Submitting is a separate step and uses prepaid credit. Review the job and spending limit in your client before submission.

BEFORE YOU RUN

A few things to know.

What is free, and what is charged?

Estimates are free. Submitted runs use your prepaid balance. Top up when you are ready. Retries stay inside the spending limit you set, never on top of it. A job that fails, or that you cancel, is billed for the resources it used — the machine time including setup, staging and the time after your command ended, storage and data transfer, and our handling of the run — never more than the spending limit you set.

Is an estimate a fixed price?

No. It is an estimate for the job you describe, not a reservation or a guarantee of availability. The charge shown after completion may change while billing is reconciled. Your total charge for the job, including retries, will not exceed your spending limit.

What happens if a run fails?

A retry or move to another machine depends on the job policy, available compute and remaining budget. Checkpoint recovery needs compatible save-and-resume logic in your code. A run may stop without completing.

Does “completed” mean my files arrived?

Check the file delivery status separately from the run status. Available output files have download links and hashes you can verify. A completed run does not by itself confirm file delivery or final billing.

What should I know about my data?

Jobs run on third-party compute. Job records and outputs may be retained after a run. Review the Privacy notice for processing, sharing and retention, and only submit code and data you have permission to use.

Can I refund unused credit?

The current Terms say top-ups are generally non-refundable, with unspent credit returned if the service closes and any mandatory consumer rights preserved. Read the full terms before adding credit.