Use Cases
See how kubbi fits into real agent workflows
Explore the patterns teams use to pass files, secrets, results, and state between agents, humans, and workflow steps.
One agent finishes the work. Another agent picks it up.
Research agents, extraction agents, planning agents, and execution agents often need to pass structured output, retrieved context, generated files, or short-lived credentials to the next step. kubbi gives the first agent a secure temporary handoff: store the payload, pass the claim URL, and let the next agent claim it only when it is ready.
Common payloads
How it works
Agent A creates a kubbi with the payload.
The workflow passes only the claim URL forward.
Agent B claims the payload and kubbi burns it on final read or expiry.
Why use kubbi?
Keep raw payloads out of prompts and tool outputs, avoid bloating orchestrator state, and make one-time handoffs easy to expire or burn after read.
A workflow step stages data for an agent to pick up later.
Orchestrated jobs, scheduled tasks, and pipeline stages regularly produce artifacts that an agent needs to act on - but the agent may not run in the same environment or at the same time. kubbi lets the upstream step store the payload and pass a claim URL through the orchestrator, queue, or task record. The agent claims the data on its own schedule.
Common payloads
How it works
A workflow step creates a kubbi with the artifact.
The claim URL travels through the orchestrator or queue.
The downstream agent claims the payload when it spins up.
Why use kubbi?
The upstream step finishes cleanly without waiting for the agent. The payload disappears on expiry or final read - no permanent storage to manage.
An agent pauses for review without keeping everything alive.
Many agent workflows need a human checkpoint for approval, QA, or domain review. kubbi lets an agent persist a draft result as a temporary handoff, record the claim URL in the task system, and exit cleanly. The reviewer claims the payload later, and the workflow resumes without depending on a long-lived runtime or shared in-memory state.
Common payloads
How it works
The agent reaches a review checkpoint and creates a kubbi.
The claim URL is added to a task, queue, or review system.
The reviewer claims the payload and the workflow continues separately.
Why use kubbi?
You get clean async review loops without dumping sensitive draft output into task comments, chat threads, or workflow state.
Workflow steps hand off intermediate artifacts without shared state.
Not every workflow step runs at the same time or in the same environment. kubbi gives orchestrated jobs, scheduled tasks, and agent pipelines a simple place to stage intermediate artifacts: one step creates a kubbi, the next step claims it later, and the payload disappears on expiry or final read.
Common payloads
How it works
A workflow step creates a kubbi with an intermediate artifact.
The orchestrator, queue, or task record carries the claim URL.
The downstream step claims the payload on its own schedule.
Why use kubbi?
It reduces schema coordination, avoids oversized messages, and gives you TTL and retrieval limits without turning temporary artifacts into permanent storage.
Other common kubbi patterns
Works with the file types most workflows use
From structured data to generated files, kubbi packages common workflow payloads behind a single claim URL.
JSON
Structured data, API responses, agent outputs
CSV / TSV
Tabular data, ETL outputs, datasets
Plain text
Logs, notes, raw output
Reports, invoices, generated documents
Images
Screenshots, charts, generated visuals
Archives
Bundled artifacts, multi-file handoffs
YAML
Config files, pipeline definitions
XML
Structured markup, SOAP payloads
Markdown
Documentation, formatted notes
Any MIME type
Binary blobs, custom formats, anything
Common payloads that travel through kubbi
{
"analysis": "complete",
"confidence": 0.97,
"next_step": "review"
}id,name,status,score 1,alpha-run,pass,0.94 2,beta-run,fail,0.31 3,gamma-run,pass,0.88
token: sk-proj-████████████████ expires: 2025-07-01T12:00:00Z scope: read-only one-time: true
stage: extraction model: gpt-4o max_tokens: 4096 output_format: json
Try kubbi on your next workflow
Create a temporary handoff, pass one claim URL, and let the next agent or human pick it up when ready.