Machine Definitions
A reusable setup for a virtual computer: its software, tools, and instructions. Teammates can start with the same setup without configuring it again.
Ticket workflows from Linear and Jira become tasks for agents in prepared Machines, with multiple issues being implemented, tested, and prepared for review in parallel. Illustrated harnesses include OpenAI Codex, Claude Code, OpenCode, Pi, and more.Your backlog.
FOR SOFTWARE TEAMS
Your backlog.
Already in motion.
Agents working in parallel.
Your team in control.
Give every issue an agent and a computer ready for the job.
Step into any task. Take the work as far as it needs to go.
01 FROM TRIGGER TO CONTINUATION
Launch a task.
Keep the environment.
A person or an event starts the task. An agent gets a prepared Machine. You can step in whenever you need to.
Follow one issue through the work, the handoff, and a new place to continue.
01
A task launches.
You start it. Or your system does.
Your next taskA request from your team
ENG-248 VISIONAST / WEB
Fix report overflow on mobile
The F&I Summary table pushes the page sideways on a phone. Keep the report and its controls inside the viewport.
The task carries the instructions.
The Machine brings the tools.
02
Work with the agent.
A whole Machine. A clear way to step in.
Web development Machine DefinitionEngineering Profile
Codex / ENG-248fix/mobile-report-overflowPREPARED
YOU
Fix the F&I Summary overflow on mobile. Keep the controls visible and let the report table scroll inside its card.
Codex in this Machine
The repository and tools are ready. Send the task when you are.
Inspect the repository components/report.tsx Test the fix in Chrome report-mobile.spec.ts
Talk to the agent as it works.
VisionAST Web app · mobile view390pxFiltersDefault filters⌄↻
DEALS
All dealerships ⌄This month ⌄
Personnel +Vehicle +F&I Performance Summary
Default configuration ⌄⚙Export ↓
| Metric | Retail | Lease | Total |
|---|---|---|---|
| UNIT TOTALS | |||
| New | 82 | 24 | 106 |
| Used | 64 | 8 | 72 |
| Total units | 146 | 32 | 178 |
| PRODUCT | |||
| Service contract | 94 | 16 | 110 |
| GAP | 58 | 10 | 68 |
| INCOME | |||
| Reserve | $42,180 | $8,620 | $50,800 |
| F&I gross | $284,700 | $53,440 | $338,140 |
| PVR | $1,950 | $1,670 | $1,900 |
Page extends beyond viewport →
Illustrated app view · sample dealership data
RepositoryNext.jsNode.jsGitChromeONE WORKING ENVIRONMENT
Steer the agent.
Use the browser whenever you need.
03
Keep going from here.
The task ends. Your next move doesn’t.
SNAPSHOT WORKFLOW PREVIEW
CAPTURED AT COMPLETION
SnapshotCode. Tools. Disk state.
LOCAL MACHINE
Pick up the environment.
On your Mac, with an agent beside you.
One task. One working environment. Yours to continue.
Choose how the task starts, then launch the example.
Interactive illustration. Snapshot continuation is a workflow preview: a new compatible Machine starts from captured disk state and applications restart. Attached Drives and Profile credentials are separate.
Let an agent
center the div.
You do not need to start a local development environment for every small fix. The agent has the tools. You get the change and the evidence to review.
FROM VERCEL’S AI SDK TEAM 75%+of issue closures in July 2026 came from its software factory.
02 FROM ONE EXPERT TO EVERYONE
What your best people teach AI.
Your whole team can use.
Your AI experts build it.
Everyone else can run it.
The right tools, skills, and tested workflows, together in a reusable Machine Definition. Ready for your team to launch with authorized access.
01FIGURE IT OUT
EXPERT WORKSPACE01 / SOURCE
MC
Maya ChenAI engineer
Her working setupFour resources. One tested workflow.
PiAgent harness↗agent-browserBrowser automation↗Revenue reportingCompany skill↗Postgres + exportsData tools↗ Tested. Reviewed. Ready to share.
One person figures it out.
02MAKE IT REUSABLE
MACHINE DEFINITION
Revenue
Operations
v4 / Built by MayaReady
MANIFESTINSTALLED
$ verify reporting-workflowChecks passed Shared setup. Repeatable launches.
A Profile supplies the login credentials and settings for the Machine.
03PUT IT TO WORK
AUTHORIZED PROFILERevenue Ops Granted to NoahReporting database / read-only role
THINAIR / MACHINE
NP
Noah ParkOperations
Revenue OperationsMachine preview
Explain last month’s revenue decline.
REVENUE INTELLIGENCE08 / 26
Revenue review
EXAMPLE OUTPUTAugust 2026 · Monthly report
Revenue change by customer segment
- Total revenue change
- −7.4%
- Active customers
- −3.1%
QUERY ATTACHED
rev_change_last_month(segment) Postgres / reporting_ro Definitions
Same workflow. Each person’s own Machine.
THE SHARED LIBRARY
Your company’s agentic resources. Together.
Explore the pieces behind the workflow.
Built to use again.
Snapshots
Save a copy of a computer’s files and installed software. Start from that copy later without setting everything up again.
Skills
Written instructions that tell AI how your company does a task. Teammates can reuse them instead of explaining the process again.
Harnesses
Software that lets AI work through a task. It sends instructions to the AI, runs the tools it requests, and returns the results.
Tools
Programs AI can use to read files, browse websites, query databases, or run code.
Conversations
A record of your requests, the AI’s responses, and your corrections. Your team can review how the work was done.
FROM WORK TO TRAINING DATA
Better workflows today.Your own training data for tomorrow.
Save the requests, corrections, tools used, and results your team reviewed. These records can become examples for testing and training AI to do your company’s work.
01Recorded work02Reviewed examples03Your training data
03 OPTIONAL SETUP HELP
Want a hand
getting started?
Use Thinair on your own. If you’d like a hand, our engineers can work with your team to connect your tools and get your first workflow running.
AN EXAMPLE FOR YOUR TEAMLess time on issues.
Less time on issues.
More time to build.
Save 2 hours on 50 issues a month, and your team gets:
- back every month
- 100 hours
- in engineering time
- $144k / year
An illustrative example at $120/hour, assuming time saved after review and workflow upkeep. Before software and setup costs.
See Vercel’s approach to issue automation Explore Early Access