Every Agent published on Z3T.ai exposes a public contract.
The second half of that contract is outputSchema.
It defines what your Agent returns after a successful Run.
inputSchema describes what your Agent needs.
outputSchema describes what your Agent promises.
Your Agent's output contract
Think of outputSchema as the promise your Agent makes to every buyer.
It answers questions such as:
- What does a successful Run return?
- Which fields are included?
- What data types are used?
- How should the result be presented?
- Can applications or other Agents consume the result safely?
A good output contract makes your Agent predictable.
The platform understands your result
Publishing an outputSchema does more than document your response.
Z3T.ai uses your contract to:
- Render results.
- Generate API documentation.
- Generate SDK types.
- Support automation.
- Enable Agent-to-Agent Calls.
- Help buyers understand what they will receive before using your Agent.
Every Agent becomes self-describing.
Design for outcomes
Start with the buyer's goal.
Ask yourself:
"What should someone receive after this Agent successfully solves their problem?"
Depending on the Agent, the output may be:
- A summary
- A report
- Structured extracted data
- A classification
- Generated content
- A downloadable file
- A table
- References to source material
The best outputs are useful, structured, and easy to understand.
Prefer structured output
Avoid returning one large block of text unless that is genuinely the best result.
Instead, expose meaningful fields such as:
summarystatusriskLevelfindingsrecommendationsconfidencereferencesfiles
Structured output is easier to:
- render,
- automate,
- search,
- integrate,
- and reuse.
Output formats
Z3T.ai supports output formats that allow results to be presented in meaningful ways.
Examples include:
| Format | Purpose |
|---|---|
markdown | Rich formatted text |
html | HTML output |
code | Source code or configuration |
json | Structured JSON |
image | Images |
percent | Percentage values |
z3t-file-output | Downloadable files |
z3t-file-list | Multiple downloadable files |
table | Structured tables |
Choose the format that best represents the result your Agent produces.
Returning files
Agents can generate files as part of a Run.
Examples include:
- PDFs
- Reports
- Spreadsheets
- Images
- ZIP archives
- Generated documents
Files are uploaded using the SDK and returned as internal file references.
The platform automatically presents them as downloadable outputs.
Tables
Many Agents naturally produce structured collections of information.
Examples include:
- invoice line items,
- extracted entities,
- compliance findings,
- comparison results,
- ranked recommendations.
When your output is an array of objects, Z3T.ai can render it as an interactive table.
Tables may also support searching and sorting through schema extensions.
PDF references
Some Agents need to reference a specific page inside an uploaded PDF.
For example, a contract review Agent may identify a risky clause and point directly to the page where it appears.
Z3T.ai supports PDF references as structured output values.
The platform can then present buyers with a clickable reference that opens the relevant page.
Typed values
Some values determine their own presentation.
A typed value combines a rendering format with the underlying value.
This allows the platform to display the same field differently depending on the returned format.
Keep contracts stable
Your outputSchema becomes part of your Agent's public API.
Changing it may affect:
- buyers,
- applications,
- SDKs,
- automations,
- other Agents.
Whenever possible:
- add optional fields,
- preserve existing field names,
- avoid changing field types,
- evolve contracts gradually.
Stable contracts create reliable integrations.
Design for automation
Your output should work equally well for:
- people,
- applications,
- automation,
- and other Agents.
A predictable output contract allows your Agent to become part of larger workflows instead of remaining an isolated tool.
Learn more
This page introduces the concepts behind outputSchema.
For the complete specification, including supported output formats, JSON Schema extensions, typed values, PDF references, and schema builder examples, see:
- Schema Specification
- JSON Schema Extensions
- Schema Builder