Artificial intelligence news.
Straight from the original announcements.

OpenRouter's AI Sales Agent Rasp Gives Its 5-Person Team Back 600 Hours a Month

October 7, 2026By the AI News Bites Editorial TeamSource: OpenRouter
OpenRouter's AI Sales Agent Rasp Gives Its 5-Person Team Back 600 Hours a Month
Image: OpenRouter

OpenRouter says its in-house AI sales agent Rasp returns about 600 hours a month to the company's 5-person sales team, while costing about $30 a day, down from nearly $800 a day for the two agents it replaced. Rasp is built on OpenRouter's Ori and handles inbound triage, first-touch emails, call briefs, notes and CRM updates.

Key points
  • OpenRouter says its Rasp agent, built on Ori, gives its 5-person sales team about 600 hours a month back.
  • Rasp sends 93% of first-touch emails on its own, and many leads get a reply in 60 seconds or less.
  • The deal cycle is 34% faster and the close rate rose 2.6x, though pricing changes and market conditions also played a role.
  • Rasp costs about $30 a day versus nearly $800 a day for the agents it replaced, with about $18 of that being inference on GLM 5.2.
  • People keep final approval on sensitive items, the calls, negotiations and relationship building.

What Rasp does and what the numbers show

According to OpenRouter, Rasp researches and qualifies every inbound lead and routes non-sales questions elsewhere. It sends first-touch emails, writes a brief before every call, drafts post-call notes from the transcript and fills in most CRM fields after each call. Sensitive items and edge cases are flagged for a person to approve.

OpenRouter reports the following results:

  • About 600 hours a month returned to the team, or roughly 140 hours a week (about 28 hours per account executive), which adds up to about 7,200 hours a year.
  • Room for roughly two more calls a day per rep at the same level of preparation.
  • 93% of first-touch emails sent by Rasp on its own, with many leads hearing back in 60 seconds or less.
  • A deal cycle that is 34% faster and a close rate up 2.6x.

The company adds a caveat on the close rate: pricing changes and market conditions moved in the same window, so not all of the gain should be credited to Rasp. It also says about 57% of the current pipeline can be attributed to capacity the team did not have before.

Where the time savings come from

OpenRouter says one demo previously required about 103 minutes of work to execute well. With Rasp, the same work takes about 44 minutes, a saving of 59 minutes per demo. Pre-demo prep dropped from 30 minutes to 5, post-demo notes from 15 minutes to 2, and CRM updates from 30 minutes to 7.

Inbound handling changed as well. Before Rasp and its predecessor Dove, sorting and answering inbound leads took the entirety of one rep's time, which pushed the team to prioritize only the very best leads. That work now takes about 25% of one rep's time. The 7% of first-touch emails that involve a human are the tricky edge cases Rasp flags proactively.

The intake form now also filters out personal email domains and routes support questions away from sales. OpenRouter says lead prioritization is fine-tuned with every new conversation as Rasp learns across the sales data.

Lessons from earlier agents and what stays manual

Rasp follows two earlier agents, Ace for account executives and Dove for business development reps. OpenRouter lists four lessons from them: run every task type through one pipeline with the task type as a parameter, put a feature flag on everything and leave it off by default, log metrics from day one, and use Slack as the interface because that is where the team works.

Some work is still done by hand. Follow-ups and re-engaging quiet deals are manual, and dropped-deal detection is under construction. Other tasks are manual by design: running the call, deciding when to push or pull back in a negotiation, building the relationship and giving final approval on anything sensitive. In OpenRouter's words, Rasp "recommends, drafts, flags, and routes" and never makes the final call on a send, a deal classification or a compliance question without a person in the loop.

Cost and model routing

OpenRouter positions automatic cost optimization as its differentiator. Its routing layer keeps searching for a cheaper model that maintains quality and swaps it in without the team doing anything. Ace and Dove were approaching $800 a day, while Rasp does the same work for about $30. The company says both figures include engineering time plus the routine outputs the agents produce.

Of the roughly $30, about $18 is inference on GLM 5.2, priced at $1.40 per 1M prompt tokens and $4.40 per 1M completion tokens as of September 2026. The company says it validated pricing for three cheaper models. The one fully listed in the source text, DeepSeek v4 Pro, is priced at $0.58 and $1.74 per 1M tokens and would bring inference to about $7.50 a day at around 200 turns a day, roughly 2.4x cheaper.

What it means for sales teams

OpenRouter says many companies are building similar agents for sales and go-to-market teams, and argues that such an agent should be easy to stand up, Slack-native, impactful and automatically cost-optimizing. It says adding a new job for Rasp is a simple chat that needs no engineer.

The figures come from OpenRouter's own team and its own product, and the company acknowledges that other factors affected the close rate. Still, the report offers a detailed look at where an AI agent can remove administrative load from a small sales team while people keep the calls, negotiations and relationships.

Related news

OpenRouter launches beta localization in eight languages
Products and toolsNovember 12, 2026

OpenRouter launches beta localization in eight languages

OpenRouter has launched beta localization in seven languages alongside English, covering Chinese, Spanish, Brazilian Portuguese, German, French, Japanese and Korean. The…

SambaNova brings Z.ai’s GLM-5.3 to SambaCloud at over 200 tokens/s
ModelsNovember 11, 2026, 19:16

SambaNova brings Z.ai’s GLM-5.3 to SambaCloud at over 200 tokens/s

SambaNova says Z.ai’s GLM-5.3 is now available on SambaCloud at over 200 tokens per second, giving developers access to an open-weight model designed for long-horizon…

OpenRouter measures the token cost of MCP, CLI tools and skills
For developersNovember 11, 2026

OpenRouter measures the token cost of MCP, CLI tools and skills

OpenRouter’s comparison of MCP, command-line tools and skills found that the way an AI agent receives a capability can dramatically change its context usage and cost. In…