Sep 7, 2026 · 37 min · 11 segments
Their best benchmark came from their own lab. That's not a weakness they hide, it's the product. Jonathan Murray, co-founder of Backboard.io, breaks down why memory (not the model) is the real…
Jonathan MurrayGuest
Rayan AliHost
Yeah, and your answer kind of connects us to the next question because I wanted you to explain Backboard.io to me.
They've got knowledge graphs, they've got, you know, file storage systems, etc.
Our view was that the speed at which things were operating, you know, we were using three or four models ourselves, you know, we were putting information into them that we were like feeling pretty uneasy about sharing.
And so when you're building a tool that's AI enabled over the last five years, what's required in that stack has just developed.
Like you got to take, you got to understand like a couple of years ago, we didn't have open claw that led to tool calling that led to, you know, autonomous agents that led to rock bot.
And we felt that if your memory could be moved portable between models, that you could operate without feeling locked in.
We wanted to give users, companies, et cetera, that feeling that they were in control.
um and so we started with memory but then we realized as new products came about we needed to build up the stack and so our our core product we call our unified api it's not just memory it starts with memory but then it has model routing it has state management it has embeddings that has parallel tool calling as adaptive context management.
And we can put that all into a container and deploy it within someone's cloud.
And you can just choose whether or not you want to use frontier models when the time comes.
And if you look at the history of software, as industries emerge, they start out fragmented.
So as people start to learn what problems actually need to be solved, they solve them, they create a tool, and you go.
Like that was a new thing that, you know, OpenRouter was one of the first two.
seven billion acquisition this past week and that's in the last couple of years so we have a model router in our platform we even have open router in our model router on our platform we have state management that's redis you've got tool a parallel tool calling that would be like a lang chain or a lang smith so all these individual tools exist we've consolidated them into a single unified architecture and so that's where our unified api sits It's really hard to demo an API.
So we then ended up building a coding harness, which is what we call our RCLI.
Yeah, and your answer kind of connects us to the next question because I wanted you to explain Backboard.io to me.
They've got knowledge graphs, they've got, you know, file storage systems, etc.
Our view was that the speed at which things were operating, you know, we were using three or four models ourselves, you know, we were putting information into them that we were like feeling pretty uneasy about sharing.
And so when you're building a tool that's AI enabled over the last five years, what's required in that stack has just developed.
Like you got to take, you got to understand like a couple of years ago, we didn't have open claw that led to tool calling that led to, you know, autonomous agents that led to rock bot.
And we felt that if your memory could be moved portable between models, that you could operate without feeling locked in.
We wanted to give users, companies, et cetera, that feeling that they were in control.
um and so we started with memory but then we realized as new products came about we needed to build up the stack and so our our core product we call our unified api it's not just memory it starts with memory but then it has model routing it has state management it has embeddings that has parallel tool calling as adaptive context management.
And we can put that all into a container and deploy it within someone's cloud.
And you can just choose whether or not you want to use frontier models when the time comes.
And if you look at the history of software, as industries emerge, they start out fragmented.
So as people start to learn what problems actually need to be solved, they solve them, they create a tool, and you go.
Like that was a new thing that, you know, OpenRouter was one of the first two.
seven billion acquisition this past week and that's in the last couple of years so we have a model router in our platform we even have open router in our model router on our platform we have state management that's redis you've got tool a parallel tool calling that would be like a lang chain or a lang smith so all these individual tools exist we've consolidated them into a single unified architecture and so that's where our unified api sits It's really hard to demo an API.
So we then ended up building a coding harness, which is what we call our RCLI.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.