Veridex vs OpenAI Agents SDK
The OpenAI Agents SDK is a provider SDK that wraps OpenAI's function-calling loop with handoffs and a UI runner. It is tightly coupled to OpenAI's models.
| Dimension | OpenAI Agents SDK | Veridex |
|---|---|---|
| Provider | OpenAI-only | Multi-provider via ModelProvider |
| Tools | OpenAI function schema | Typed contract + safety class + sandbox |
| Policy | None | Engine + packs |
| Memory | Minimal | Tiered + reconciler |
| Approvals | None | First-class |
| Checkpoints | Run-thread (OpenAI-side) | Differential, durable, portable |
| Audit | OpenAI trace UI | Hash-chained, signed, portable |
| Adversarial proofs | None | Built-in |
| Self-host | No (OpenAI infra) | Yes |
Compose
A Veridex agent can call an OpenAI Assistant as a tool, or an Assistant can hand off to Veridex via the adapter:
import { handoffToVeridex } from '@veridex/agents-adapters/openai';
case 'process_payment':
return handoffToVeridex({ agent: treasuryAgent, input: args.userIntent });When to use the OpenAI SDK
- OpenAI-only, simple workflows.
- Prototypes that benefit from the Assistants UI.
When to use Veridex
- Multi-provider (OpenAI + Anthropic + open-source models).
- Compliance requirements.
- Multi-tenant.
- Adversarial threats.