> ## Documentation Index
> Fetch the complete documentation index at: https://docs.portkey.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# July

July brings cross-API routing between Messages and Responses API, expanded agentic client support in Prisma AIRS, and a soft deny mode for guardrails that keeps sessions alive for agentic clients.

We've also added Lightning AI and Akto as new integrations, and shipped improvements across Anthropic, Vertex AI, and Bedrock.

See what's new:

## Summary

| Area | Key highlights |
| :- | :- |
| **Gateway** | • Messages API to Responses API routing |
| **Guardrails** | • Prisma AIRS: agentic client support <br /> • Soft deny mode (HTTP 200) <br /> • Akto |
| **Models and providers** | • Lightning AI <br /> • MCP user attribution in logs |

## Gateway

### Messages API to Responses API Routing

Route Messages API requests to Responses API models—like OpenAI's Codex—without changing your client code. Requests and responses are transformed automatically in both directions.

Learn more: [Responses API](https://portkey.ai/docs/api-reference/inference-api/responses/responses)

## Guardrails

### Prisma AIRS: Agentic Client Support

False positives drop significantly for agentic clients like Claude Code, Cursor, and Cline. **`strip_scaffolding`** removes agent-harness content—system prompts, MCP tool instructions—before scanning, so Prisma AIRS inspects actual user content. **`scan_scope`** gives you control over which messages get scanned.

Learn more: [Prisma AIRS](https://portkey.ai/docs/integrations/guardrails/palo-alto-panw-prisma)

### Guardrails: Soft Deny

Guardrail denials no longer crash agentic sessions. With **`soft_deny_200`**, a denial comes back as HTTP 200 formatted as a chat completion response—so clients like Claude Code that treat 4xx as fatal errors keep running and see the denial as a model response.

Learn more: [Guardrails](https://portkey.ai/docs/product/guardrails)

## Models and providers

### New providers

* **Lightning AI** — chat completions

### MCP Gateway: user attribution in logs

Every MCP request now traces back to a specific user—derived from OAuth claims or the API key name—for clearer per-user visibility in analytics.

Learn more: [MCP observability]()

## Bug fixes and improvements

* Vertex AI: Qwen and xAI Model Garden routing added
* Workers AI: reasoning support for K2.7 and GLM-5.2
* Anthropic: cost attribution now differentiates batch, fast, and standard pricing tiers

## Support

<CardGroup cols={2}>
  <Card title="Need Help?" icon="bug" href="https://github.com/Portkey-AI/gateway/issues">
    Open an issue on GitHub
  </Card>

  <Card title="Join Us" icon="discord" href="https://portkey.wiki/community">
    Get support in our Discord
  </Card>
</CardGroup>


## Related topics

- [July](/docs/changelog/2025/july.md)
- [Enterprise Gateway](/docs/changelog/enterprise.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.