This AI startup helps developers safely and cheaply build on top of LLMs from OpenAI and Anthropic. Read its pitch deck.

• Requesty wants to act as a gateway between developers and multiple AI providers.
• The startup says its tech tightens security and brings down the costs of accessing large language models.
• Requesty raised a $3 million seed round, led by the venture capital firm 20VC.

Leaders across industries are encouraging employees to increase their AI usage — but it’s becoming costly, and there are risks as people upload sensitive information into an array of large language models.

Requesty wants to help companies bring down the expense and mitigate against any AI perils.

Originally founded as a data analytics startup, London-based Requesty pivoted this year to focus on infrastructure. Its technology sits as a gateway between AI providers like OpenAI, Anthropic, and Grok, and developers who are building AI-powered applications for tasks like coding, booking travel, scanning invoices, and translation.

Thibault Jaigu, Requesty’s cofounder, describes the company as a kind of Cloudflare — an internet infrastructure and security company — for the AI era. Requesty wants to act as the central pillar connecting different AI servers.

A typical pain point for companies is that engineers have different preferences for the AIs they want to use. They often configure several different models without centralized governance or control around their usage.

“The admins — which is IT security or DevOps — don’t understand what the engineers are using, why they’re using it that way, and what the potential risks are,” Jaigu said.

Risks could include sending customer and other proprietary data to the AI or not revoking a developer’s access key after they leave the company.

Requesty aims to centralize these controls — such as spending limits and API access — across multiple AIs.

Requesty says another benefit of having a unified system is that it can help developer teams better optimize between different providers. If OpenAI is down one day, for example, it can route requests through Anthropic instead.

Requesty also says it can save companies money by using a technique called “prompt caching,” which stores frequently used prompts and reuses them.

In the four months since its pivot, Requesty said its platform is being used by more than 25,000 developers at AI app-building companies, including Mozart AI and Blue Morpho. The company said it has passed $1.5 million in annual recurring revenue.

AI gateway platforms are a newly emerging category in the startup space, creating an infrastructure layer between AI providers and the developers who build on them. Other similar startups in the space include Humanloop, whose cofounders and senior team were acqui-hired by Anthropic last month.

Requesty announced on Friday that it had raised a $3 million seed investment round, led by 20VC and with participation from venture capital firms Tapestry VC, Insiders Ventures, and Tiny Supercomputer.

Jaigu said Requesty intends to use the funds to staff up. It has a head count of five people that it plans to grow to around 15 to 20 next year, hiring primarily across its engineering and growth teams.

Check out the pitch deck Requesty used to secure its $3 million seed investment, shared exclusively with Business Insider. Some of the slides have been omitted or redacted.

Requesty’s tagline: ‘The developer’s gateway to safe AI’

Its cofounders have recent experience at companies including Amplitude, TikTok, and Palo Alto Networks

This slide demonstrates some of the current challenges within companies’ AI setups

The slide says those include a lack of unified governance, gaps in security and compliance, limited visibility and cost control, a poor developer experience, and exposing companies to vulnerabilities like outages.

This slide demonstrates the return on investment Requesty says it can generate for a company spending $1 million on its tech

The savings come from techniques like prompt caching, automating audits, boosting productivity, and mitigating risks.

Requesty describes its four-step process

First, companies centralize all their AI from different departments into one. Then, they add controls like spending or model limits. The third is building in observability and cost optimizations, such as performing a task using a cheaper AI model. The last step is safety, such as building a rule that no PDFs from the company are uploaded to the AI.

This slide demonstrates the typical Requesty workflow

A developer might use a chatbot or coding assistant. Before their request is sent to the AI provider, Requesty would analyze, for example, whether the request is from a valid user with the correct permission, or whether it could be routed through a faster server in a different region.

Requesty outlines three major growth areas: control and governance, visibility and accountability, and flexibility and optimization

The final slide takes an in-depth look at Requesty’s features and their benefits

Read the original article on Business Insider

The post This AI startup helps developers safely and cheaply build on top of LLMs from OpenAI and Anthropic. Read its pitch deck. appeared first on Business Insider.