OpenRouter has integrated an AI agent named Rasp into its five-person sales team. Built on the Ori engine, the agent handles lead qualification, initial client outreach, meeting briefings, call transcript summaries, and CRM synchronization. According to head of sales Michael Haimovitz, this automation saves the team approximately 600 hours per month.
The agent reduces the time required to process a demo call from 103 to 44 minutes by automating pre-meeting research and CRM updates. This allows each manager to conduct two additional calls daily. Rasp manages 93% of initial client emails, delivering responses in under a minute, while flagging complex cases for human review.
The company reported a 34% acceleration in the deal closing cycle and a 2.6x increase in conversion rates over several months. Additionally, 57% of the current sales pipeline is now attributed to capacity that was previously unavailable. While market conditions and pricing changes played a role, the company notes a clear correlation with the agent's implementation.
The operational cost of the agent has dropped significantly. While previous versions, Ace and Dove, cost around $800 per day, Rasp performs the same workload for approximately $30. Of this, $18 is allocated to the inference of the GLM 5.2 model. The Ori routing layer achieves this by selecting cost-effective models for routine tasks.
The company aims to further reduce daily inference costs to $0.50 by migrating standard operations to even more affordable models. Despite the high level of automation, the architecture ensures that human oversight remains mandatory for final decisions regarding deal classification, compliance, and final communication approval.
