Getting tool call accuracy right is key for a smooth Agent UX. In our latest benchmarking post (link in the comments), we break down how adding more context or tools to your prompts can actually make accuracy drop from 73 percent to 66 percent.
Want to keep your agents sharp? Check out this quick demo on how to set up continuous evaluation using Maxim AI.
Ready to level up your agents? See how Maxim can help you build high-quality, reliable agents that deliver real results - https://evals.run
Maxim AI
Hello PH community, I am Akshay from Maxim, and today we’re excited to officially announce the launch of Bifrost, a blazing-fast LLM gateway built for scale.
What is it?
Bifrost is the fastest, fully open-source LLM gateway that takes <30 seconds to set up. Written in pure Go (A+ code quality report), it is a product of deep engineering focus with performance optimized at every level of the architecture. It supports 1000+ models across providers via a single API.
What are the key features?
Robust governance: Rotate and manage API keys efficiently with weighted distribution, ensuring responsible and efficient use of models across multiple teams
Plugin first architecture: No callback hell, simple addition/creation of custom plugins
MCP integration: Built-in Model Context Protocol (MCP) support for external tool integration and execution
The best part? It plugs in seamlessly with Maxim, giving end-to-end observability, governance, and evals empowering AI teams -- from start-ups to enterprises -- to ship AI products with the reliability and speed required for real-world use.
Why now?
At Maxim, our internal experiments with multiple gateways for our production use cases quickly exposed scale as a bottleneck. And we weren’t alone. Fast-moving AI teams echoed the same frustration – LLM gateway speed and scalability were key pain points. They valued flexibility and speed, but not at the cost of efficiency at scale.
That’s why we built Bifrost—a high-performance, fully self-hosted LLM gateway that delivers on all fronts. With just 11μs overhead at 5,000 RPS, it's 40x faster than LiteLLM.
We benchmarked it against leading LLM gateways - here’s the report.
How to get started?
You can get started today at getmaxim.ai/bifrost and join the discussion on Bifrost Discord. If you have any other questions, feel free to reach out to us at contact@getmaxim.ai.
@akshay_deo Wow, this is super cool, looking forward to using it! Congrats on the launch!
Maxim AI
Hi, I’m Pratham.
I’ve been building products for a while now, and over time I’ve become deeply invested in backend systems that don’t just work, they scale, stay lean, and never get in your way. That’s the philosophy behind Bifrost, the open-source LLM gateway we’ve been building in Go.
Here’s what we focused on:
Architecture-first — so adding features never compromises performance.
Go, done right — full use of its concurrency and memory optimization features.
Lightweight core — with a powerful plugin system to toggle features like switches.
Multi-transport native — HTTP, gRPC(planned), and more coming in.
The result? A self-hosted gateway with ~11µs mean overhead at 5K RPS, support for every major LLM provider, built-in monitoring, hot-reloadable config, governance controls, and a clean UI built for production from day one.
You can get started here: getmaxim.ai/bifrost
Join the Discord to geek out with us: getmax.im/bifrost-discord
AltPage.ai
Whoa, love seeing a blazing-fast LLM gateway! Juggling slow API calls has been a pain for my side projects—can’t wait to see how much Bifrost speeds things up.
Maxim AI
@joey_zhu_seopage_ai Thanks! We’d love to hear how it performs in your setup.