- URL or base64 images
- 524,288-token context
- OpenAI message format
Capabilities
ERP-1
One model for long-running work.
ERP-1 combines a large working context with multimodal input, reasoned text output, and function calling for agents that need to act across business systems.
- Native server-sent events
- Final usage metadata
- Up to 65,536 output tokens
- Automatic tool choice
- Long tool histories
- Structured function arguments
Inference data policy
Provider-controlled systems
Content is processed, not collected.
ERP-1 operates with zero retention for inference content. A routing platform or customer application may apply its own independent policy.
No prompt or completion retention
Messages, images, tools, tool results, and generated text are processed transiently and are not written to provider application logs.
No training on customer traffic
Inference prompts and completions are not used for pre-training, fine-tuning, evaluation datasets, advertising, or human review.
Limited service metadata
Request identifiers, timestamps, routes, status, latency, usage, and security metadata may be retained for reliability, abuse prevention, reconciliation, billing, and legal obligations.
United States edge, Japan inference
Traffic is encrypted in transit. The public edge is operated in the United States and the current GPU deployment serves inference from Japan.
ERP-1 is declared to routing partners as zero data retention and not HIPAA compliant. See the ERP.AI Privacy Policy and Terms of Service for the company-wide legal terms.
ERP-1
Operated by ERP.AI
Bring ERP-1 into your agent workflow.
Talk to ERP.AI about model access, capacity, and commercial terms.