The neutral runtime for your agents
Wodex Runtime runs the open-source agent harness behind OpenAI Codex - on Wodex cloud or inside your boundary. Any model routes through the gateway, executors live wherever your code lives, and sessions and audit logs stay in your tenant.
The four walls of the hosted agent cloud
Every claim about OpenAI's Agents API below links to OpenAI's own documentation.
| OpenAI Agents API | Wodex Runtime | |
|---|---|---|
| Model choice | OpenAI models only docs ↗ | Any model via the gateway - OpenAI, Anthropic, DeepSeek, Qwen, open weights |
| Harness location | Always on OpenAI's cloud - "OpenAI runs the agent harness" docs ↗ | Wodex cloud, or inside your boundary (private deployment) |
| Data residency | /v1/agents is not in the data-residency program docs ↗ | Sessions and audit logs stay in your tenant region |
| Retention | Not eligible for Zero Data Retention; session state kept until deleted docs ↗ | Nothing retained beyond operations - our policy, not a certification claim |
| Sandbox billing | Container rates, billed per minute docs ↗ | Priced per agent seat, not per container-minute |
How it works
A resident harness, executors anywhere, any model.
The harness stays resident
The open-source agent engine runs on Wodex cloud or inside your boundary. Sessions, compaction, and recovery are managed; state and audit stay in your tenant.
Executors live anywhere
The agent's hands run on laptops, CI runners, VMs, or your VPC. Outbound-only connections - no open ports, no network changes.
Any model, per task
Every call routes through the Wodex gateway: OpenAI, Anthropic, DeepSeek, Qwen, or open weights, switched per task with fallbacks.
One console for human and agent seats
The same admin that governs your team's ChatGPT seats governs agent seats: quotas, audit, skill distribution, and model routing - one ledger for both.
FAQ
How is this different from OpenAI's Agents API?+
OpenAI manages the harness on its own cloud, accepts only OpenAI models, keeps session state until you delete it, and bills hosted sandboxes at container rates. Wodex Runtime runs the same open-source harness on Wodex cloud or inside your boundary, routes any model through the gateway, keeps sessions and audit in your tenant, and charges per agent seat.
Where do sessions and logs live?+
In your tenant. With Wodex cloud hosting they stay in your tenant space; with private deployment they never leave your boundary.
Do you hold compliance certifications?+
Not yet, and this page makes no certification claims. Our policy is that nothing is retained beyond what operations require - we would rather show you the architecture than a badge.
How is it priced?+
Per agent seat, not per container-minute. Pricing is published at launch. Email [email protected] for early access.
Which models can agents use?+
Any model the Wodex gateway routes: OpenAI, Anthropic, DeepSeek, Qwen, or open-weight models, switched per task.
Put your agents on a neutral runtime
Tell us your team size and workload, and we will line up early access in order.
