Name: Seqera Co-Scientist: Intelligent Bioinformatics That Scales Across Your Organization
Uploaded: 2026-05-27T19:01:27.740Z
Duration: 57 min 12 s
Description: Seqera Co-Scientist: Intelligent Bioinformatics That Scales Across Your Organization

Transcript for "Seqera Co-Scientist: Intelligent Bioinformatics That Scales Across Your Organization": Alright. Welcome, everyone, to the coscientist webinar. We're gonna give it a few more minutes for other folks to trickle in. In the meanwhile, you should see a poll available now in the poll tab. It'd be great for us to get a better understanding around what tools you're using today, AI, all on tools for your bioinformatics work. Some general housekeeping, should you have any questions, feel free to ask them in the q and a. We will be doing a q and a live towards the end. Awesome. It looks like we have some introductions in the chat. We'd love to hear, you know, who you are, what your role is, or maybe also kind of how you're using AI today in bioinformatics work. Alright. So I'll let the rest of the folks trickle in. Please feel free to do your introductions, but I think we will go ahead and get started. So welcome, everyone. Thank you for taking the time today to learn more about Secure Co Scientist. Today's webinar will be really focused on using Co Scientist as kind of your intelligent bioinformatics agent that scales across your organization. To quickly start with some intros, I'll introduce myself. My name is Isha. I lead the scientific engagement team here at Secura. We're also joined today by my colleague, Sasha. Sasha, do you wanna intro yourself really quickly? Hi, everybody. My name is Sasha DeGaiev. I'm the head of AI here at Secara. Really excited to show you guys kind of some of the things that we've been building over the past couple of months. Alright. Thanks, Sasha. So before we start talking about coscientist, I want to walk you through some background about Nextflow and Secura. Nextflow is the open source workflow language that's become the de facto standard for modern bioinformatics. As data volumes exploded, reproducibility became nonnegotiable. The industry required a standard, and Nexlow has become that. You'll see this in the chart kind of depicted on the left side. Citations tracked against Nexlow compared to other workflow language have pulled meaningfully ahead over the last few years. And the numbers on the right are also kind of really what matter for this conversation. Nexo is in every one of the top 20 pharma companies and organizations. There are over a 100,000 Nexo workflow runs happening every single day globally, and across an estimated of 40% of all bind modern bioinformatics workflows use Nexlo. This matters today because Nexlow is the substrate for everything that is being built upon, including and increasingly AI agents. So every claim we'll make later about reproducibility, auditability, and reliability of agentic systems starts from the fact that the underlying pipelines are written in Nexo. And here's a slightly different cut of that same story. What you're looking at is the daily Nexla runs over the last three years. The dotted line marks early twenty twenty five, which is when we've started seeing kind of meaningful adoption of generative AI and LLMs in how scientific teams work in early twenty twenty five. And you can see that curve change shape slowly. Daily runs are going from a baseline to tens of thousands to now over spikes, of around 500,000 runs, and the rolling average is doubling. And the reason this matters is that AI is allowing us to do science and generate science a lot faster than we could as humans just alone. Agents propose more experiments, more analyses, more iterations. And to do that, they all need to actually run reproducibly on real data and real compute. And Nexla is what's absorbing that volume. The volume you see here is also what drove Secura to kind of build an organizational layer on top. And to quickly kinda describe who's behind Nextflow, it is Secura. Secura is the company that builds and maintains Nextflow, but adds kind of the Secura platform around it. We work with about a 150,000 scientists globally, 14 of the top 20 pharma companies, and over a 130 customers, some of which the logos have been shown at the bottom below. And we also, last year, raised a 26,000,000 series b, specifically to invest in AI in the agentic side of the platform, which is what essentially brings us here today. This is kind of Secura platform, your trusted platform for bioinformatics at scale. And at the foundation of it is Nexlow, your kind of workflow engine power powering your scientific pipelines. But the layer across and on top of it is also the foundation that makes sure that you have standardized workflows, standardized execution of your pipelines, and infrastructure as well. And in the middle is the pillars for how Teams actually work, your pipelines, but the data that you're actually analyzing and interactive analysis as well as any kind of newer AI and ML work underneath being supported by any and any kind of major compute back end that you might need. The reason I'm showing you this today is before we go into Coscientists, we want to emphasize that AI isn't just bolt on. It's built into Secura platform. On top of Nexlo, the same auditability, reproducibility, and same security posture. This all carries through. So when you adopt Co scientist, you're not just getting AI, you're getting an execution platform that happens to have AI built into it. And AI is powerful, but also very unpredictable as we all might know. Your AI agents are only as good as the infrastructure underneath them. So whether you're building AI agents, custom dashboards, or terminal workflows, Secura platform's AI first architecture gives you the flexibility to make sure you can integrate into however you work best. So this question of what ran with what parameters on what data and can I prove it gets a lot harder to do with AI? Your pipeline needs to make sure it meets FDA validation requirements. Your data sometimes cannot leave specific regions or environments, and you need to ensure to have complete audit trail for your runs. And that's what Sakara essentially gives you from day one, an API first, platform by design, a system of record for every run, every result, every parameter, configurable compute environments for any environment where you are running your analyses, whether it's cloud, hybrid, or on prem, and really important things for kind of enterprise organizations for authentication, role based access controls, workspace isolation, and also ensuring that you can still use your pipelines as they're built where they're built through get back to reproducibility, data lineage to make sure that you can trace any result back to the exact inputs, and security and compliance to make sure that if you choose to do kind of GXP, analysis and work that you have aligned audit trails and roll scope data access. So what I'll do now is pass it over to my colleague, Sasha, to kind of talk a little bit more in-depth about Securicosigntest. Hi, everybody. My name is Sasha as I introduced myself, earlier. I'm the head of AI here. For all of those that are joining us late, you're in time for the demo, so now is a great time to to join. The so what first, I wanted to talk about is what Secara Coscientist is. So, oh, I'm seeing people cannot hear me. Can folks hear me, in the chat? Yeah? Okay. Okay. Great. Okay. So the, SCARECO scientist is an agent, that helps you build, launch, and automate, a lot of your bioinformatics work. And how this is, accessible is through three this agent is accessible through three different mediums. Number one is the web chat. This is a, web interface that, is very reminiscent for those of you that let enjoy using, systems like ChatterGPT and and, Claude. Then there's the CLI, which is very, which is for a lot of of, advanced developers that want to be, developing pipelines and want some of that local context. And then finally, there's the scientist view, which is a simplified view for what lab researchers that, may not be as comfortable with bioinformatics, but want to launch pipelines and want to have a place for self-service. Right? So that means that kickoff pipelines, ask basic questions about what happens on specific runs, and fix any issues. So first up, I wanna kinda give you guys a demo on how do we build with coscientists and what it can do today. And for that, I'm going to switch over and do a little screen share, and we'll start with the web product. Okay. Great. So I'm just going to stop sharing and then share my screen. Alright. Can everybody see my screen? Yes. Alright. Cool. So this is the Sakara, coscientist, view. The major way and the right way to think of this is it's a great place to hand off some of the bioinformatics tasks that you may have. So in here, the first one we gonna do is we're gonna select the, repository where we want to work and we wanna start with an ambitious task. This, this repository here, the demo, repository is a basic RNA sequencing pipeline. And for us, we wanna swap out the aligner that we're using within this repository. So for us, what we wanna can do here is we can say, hey. Can you please log out the aligner from salmon to star in this repository? And here, we've submitted our first kind of work request into the agent session. Now, the major difference between, legacy systems and our system is that it doesn't just it doesn't just work on one task at a time. It can work on several tasks simultaneously. So here, I'm gonna start a new task and I'm gonna do something different. I'm going to go ahead and I'm going to pull up an old run. I'm gonna say, hey. Can you please tell me what happened on this run? And again, here, I've submitted it, and you can see that both the requests are getting worked on here. And just for, completeness sake, we're going to ask about a completed run this time. And we're gonna say, hey, can you please summarize the reports that are associated with this one. And so here, what we now see is we see a submitted status, and now we see a all of these tasks starting to get executed at the same time. Now the power of coscientist is that it is able to kind of reach in and get information about every single thing that happens within the secure platform. In here, we pulled in information for a specific run, and now we can ask specifics about it. You can say, Hey, can you please tell me about the QC on this run. Look at the multi QC report. And so we can see kind of like this isn't working now. We see our first run has completed here. And so this was a, a run that failed on the Secura platform. And here, we can see that, what Secura Cosciences did was that it retrieved is that it retrieved the log information from it. Then it looked at the Nexo log, and then it reached in and properly, debugged that this was an error where we, we got an error with spot instance, reclamation on AWS. Right? And so what we can do now is that using the same system, we can resume the run or we can consider, swapping out the Nexo config within it to try to get it to do something different. Now, the first task that we asked for it to do is we asked it to, swap out the aligner for seven and two star in specific repository. And so, again, what we can see here is that the system was able to, create its own branch, make its own edits, and then to create its own PR, which swaps out, an aligner within within the, within this repository. Right? Now the key the key thing that, it does in here though, that we want to, that we really want to highlight is that it does all of this using, using a task based management system that allows it to stay on course and use approved modules from the NF core library in order to create this specific pull request. Within the web interface, you can use this to queue up a bunch of your bioinformatics tasks and or, questions to very, very quickly get done with all of the tasks that you may have. Finally, we kind of jump to our, jump to our, multi QC work. We can see that the system is right now, downloading the, downloading the report and it is going ahead and reading it, reading it in right now. If we scroll up here, we can see the MultiQC report is pulled here. And then you can review all the information that is in here. Right? And so, again, what what cosigness is really good at is it's really good at getting being connected to all of your all of your Secura systems and letting you do everything end to end. Right? Whether that be creating EPRs and code submissions, reviewing runs that have completed, and fixing the runs that are on the, that are on the platform. Now this is all running on the on, your exact compute. So one of the things that you will see in the bottom here is this sandbox compute here. Now for enterprise clients that are especially on AWS, this sandbox is going to be entirely within your, it's going to be entirely within your infrastructure. So what that means is that everything is being run, you know, entirely within within your within your system. So meaning that the cosign test is using your inference endpoints and it's using your compute to perform any of the calculations that it may need in order to get done what it needs to get done. And so it looks like it's still running the QC reporting here, and it's finally, lining up. And so what it's telling us is that all of the, all of the pipelines, all sorry. All of the samples seem to have very good mapping rates, but it looks like that there's some duplication in some of the samples here. Again, all of this is possible within cosign to us due to the sandbox that it has access to. And again, all of this runs all on your own, all on your own, compute. Right? So again, this is the this is the power of coscientists to handle a lot of the building tasks that you may, that you may have. Now for more complicated work, and let me go ahead and share my screen for, for the local experience. What we have is we have a CLI, option where, essentially, you have the similar, a similar, level of of, control but within your local, with within your local environment. And so in here, what we have is we have a a snake make pipeline that is very, very basic, but we want to convert it to next level. And so these are some of the things where cosigness does a really good job of doing this for you. So you can do things like, hey. Can you please convert this pipeline to Nextflow? And, you have all of the all of the, kind of, built in, agent harness of cosigness, but locally on your on on your machine. Now these systems run, you know, for, thousands and thousands of tool calls. So they go, they go very, very, very, for for very sorry. Let me just zoom in here. They go for a very, very long time in order to be able to complete any of the, any of the tasks that that you may have. Now what you're seeing here is that cosigness can go ahead and check a lot of the, of the existing code and then generates a kind of a plan here. And then it's going to start doing a lot of the work. Now we're not gonna sit here and watch it do this exact, the exact workflow end to end. But trust me, these are the systems that are available both on the web and on the local side. So that way, you can go ahead and kind of do whatever it is that you need to you need to do in order to, complete your your bioinformatics tasks. So this is what we this is the, the build suite with what we have. But before we go, I do want to kind of talk you guys through a specific example that we had with an RNA sequencing pipeline, very, very recently. And I'll jump back into the slides. Okay. So, the, I wanted to kind of, talk through, you know, a workflow like the one that I had just shown the start of, but with a a pipeline that we kind of had to fix internally. So we had somebody come in who had a RNA sequencing pipeline that was written in batch, that was fairly basic, and they used our cosign to this product to move it into, into Nextflow for cloud orchestration. Now what this look like is essentially a looped environment where cosigness kept making, you know, commits, kept making changes, kept kind of iterating and iterating and iterating until it, until it kind of succeeded. Right? And so for us in this specific use case, it took about 84 iterations of the agent kind of iterating on itself until it got the right, until it got the right, the right, byte for byte match with the task that it that it had. Right? And for where Cosigncist really, really shines is that the system is set up for to be connected with Secara so that it can make changes to the code, push them up, and then run them again. That way it can make sure that everything that it's doing is actually, you know, making sense and that the outputs are one to one. So what it kind of of, the the, things that Cosigness was able to pick up, number one is that it was able to, update some of the tool versions. Two is that it obviously upgraded the tool the, orchestration. And then it made the system kind of be much more kind of cloud native and be able to use things like AWS batch for the running of this pipeline. So in all in all, you know, the, the pipeline was able to be kind of lifted and modernized into a next level context, all all with coscientists, and all reproducible and all with the power of, you know, that looped environment where coscientists on its own was able to get 84 commits, you know, end to end, to get to where it needed to get to. Now all of this is really, you know, all of this works really, really well, because Co Scientist is natively hooked up to the Nextflow registry, which is a collection of common n f core modules that are battle tested in n f core and, Sarek, all of those, you know, pipelines that we know and love, and you should be reusing. So when cosigness needs to rewrite a process or needs to make changes, what it does first is that it looks to the Nexo registry to make sure that everything is being done, is being done correctly. Now the concept of a registry is very, very interesting because, you know, what we don't want these systems to be doing is we don't want them to be generating de novo code every single time. We want them to be using reusable code as much as possible. Now this concept, we have also for private code as well. Right? So if you guys if you guys if anyone on the call wants to bring coscientists into their organization, coscientists natively will use the NF core registry. But there's also the ability to build and maintain a private registry with some of your private n f core, Nextflow, processes and plugins. So that way, you know, you can customize your specific, your specific cosigness experience for your code. Now the concept of having your own registry was, you know, is not new. We've had the ability to self host the your own registry for a long time. But most people did not love the idea of a registry, but did not love the idea of maintaining that registry. And this is kind of one of those instances where the technology is really meeting the moment at the right time because the you you know, we're using coscientists as a way as an agent to manage this registry. So what it does is that it, reviews your existing code base for best practices and then it promotes certain bits of the code to your module library for you to approve. And then it also can refactor legacy code with some of the new with some of the, you know, bits of code that you have promoted. So you can think of it as a as something that continuously improves your code quality. Now all of this is obviously, shareable. You know, we know that a lot of people on the call probably use a system like codex or Cloud Code. And we wanna make the investments that you guys have made into those tools even better. Right? And so what we have is we have the Secura MCP, which provides you access to not just the Nextiva module library, but also your private, your private module library as a way to one, look for bits of reusable code, but two, as a way to use it, to improve the code that it is generating de novo. So that way, you know, you get less hallucination and more trusted code. Now, obviously, like all of this, we're we are really happy with coscientists and the build step is that is in the quality of the tokens that we produce. On our internal, benchmarks and our external benchmarks, we're seeing, you know, about a 2020% relative lift, against the competition, but at a dramatically lower token cost. So we you can use our cosign to suite not only to work on the your pipelines directly, but you can also burn a lot of this benefit to, the tools that you're already using. Right? And all of this is powered by the Secera ecosystem and the Secera MCP. So the way the best way to think about it, right, is that coscientist maintains a you know, is really is your bioinformatics specialist that also builds a module library that so that way all of your other agents that either you or other members of your team are using every day to make sure that the code quality within your organization stays very high. So that was the build step. Now on the launch step, essentially there's now three ways to launch a pipeline. Number one is through the web chat that and the CLI that I have, started off with, but also with the scientist view. So first up, I'm gonna, talk about running the pipeline with the chat interface. The best, the best thing that the chat interface does is that it it transforms natural language into specific pipeline parameters that a user may have. For example, a typical, you know, the typical NF core RNA C pipeline has around a 170, configurable parameters. It's a lot for anybody, even an expert to, to, work with. Now where we're really good at that is taking the natural language and kind of powering these powering the, ability to launch really complicated pipelines just through just through that. Now the CLI where that's really good is that it allows you to enter into a looped environment where essentially you can use the CLI to immediately test the code that you are publishing. So the CLI can not only edit the code, but it can promote it to a specific branch. And then you can use the Secura platform to launch it immediately and to test the changes that it is making. Finally, there's the scientist view. And what the scientist view is is that it's a very simplified version of the secure platform that essentially has three buttons. Launch button, which allows a user to launch the launch a pipeline. Two is an explain button, which launches the user into cosigness where a user can, reason over results. And three is a fix it button, which allows the user to simply fix the a broken pipeline with, you know, with co scientists right away. Now the most exciting part of so that's kind of like where we are today. Now the most exciting part of the conversation is where we're going. Right? This right now, what I'm gonna show you guys is a lot on the automation. This is what's in the current preview environments right now and what I wanna show you in terms of what it where things are going. So the where we see, where we see bioinformatics moving is less from this artisanal science that it is today where, you know, a sign of samples or an email or something comes in and it, buying competition does something and much more to a factory model where you have events from the outside world, whether that be new samples, new lens events, and sequencer events that are then augmented by context and rules that feed into some agent system that then produces an output. Right? And so this is what we see as the kind of the developing trend into where bioinformatics is going and what we want to build the future of secure around. So what this means is essentially a set of triggers and a set of responses that the system's gonna be able to make. And these triggers can be either bucket events or pipeline statuses and these can be responses can either be deterministic or, or agentic. Now for most, for most use cases, you know, we want to be as deterministic as possible. So when new data arrives, for example, you know, a sequencer, drops to complete a run, we know exactly the pipeline and we know exactly what we're gonna do. So we don't really even want to use AI. We wanna use deterministic actions to launch a pipeline. That is, you know, the first level of automation that we want to make dramatically easier on the new platform. Number two is failure recovery. So when a pipeline today fails, what we're gonna able what we're gonna allow folks to do is make it dramatically easier to automatically start up a cosign to a session that automatically looks at the errors and then resolves and relaunches the pipeline. And then finally, we have result interpretation, which is going to be for a first look at completed runs. So if you have something that, launched, completed successfully, what we wanna do is we wanna move into a world where cosigness does the first look and gives you some idea of what of what to do next on on a specific set of results. And so, the the way that we we want to do this all is with the appropriate permissions. Right? And so what this means is that the agents have their own permission scheme, meaning that you can select, and make sure that things are, you know, that the agents don't get overeager and that you're not in a doom loop. Two is that all the code changes follow your code practices and the agents never commit commit to your main or master branches that are highly sensitive. Finally, there's an audit or lineage log for each agent action to make sure that whatever it has happened, you can actually tell either yourself, your team, your supervisor exactly what the agent did. Right? And so right now, I wanna kinda show you guys give you guys a preview demo of where cosign to us is, is going. So one second. Just gonna All right. Can everybody see my screen? Yes. Awesome. Essentially what we want to do is give the user an ability to add a new action. Essentially what we want to do is we want to do something like, please fix a pipeline. And what we wanna do is we wanna say, hey. Every time a pipeline failed that is the n f core r n a c pipeline, we wanna trigger an agent. We wanna tell the agent, please fix this pipeline. We want to select a specific agent and let's specifically select our debug agent. Now in here, you'll see that this agent has a particular set of permissions, particular pipeline launch budget, and a specific set of computing environments that it has access to. Additionally, you'll see the token budget that it has, number of repositories, and the resources that it it has access to. Then we can go ahead and add that, and that will be a save action that takes place every single time an N of Core RNA seek pipeline, fails. But let's get more specific. Right? Let's go ahead and see exactly what this debug agent is all about. So what we do is we go ahead and we click on the debug, debug agent, and we see this menu here. Okay? Looks pretty good, but we wanna see exactly how it was set up. So we click on the edit, and this is where we see the configurable parts of the agent. Number one is that we see the identity. So this is where you define the agent system forms and its description. Two, and this part is the really important part. This is the permissions and budgets. So this is where you can kind of select and say what you want the agent to be able to do. So because this is our debug agent, we wanted to have the ability to launch pipelines. We wanted to have the ability to, read data and wanted to have the ability to access compute environments. Additionally, we wanna give it an itemized cap of of, the number of tokens that it can use and number of pipelines that it can, that it can launch. And for the pipelines it can launch, we want to provide a max number of CPUs and memory that its pipelines can take. This is really meant to make sure that we don't run out of, compute space and the agents kind of stay more or less on the rails. Especially for users that are new to automation, you know, we want folks to feel comfortable when they set things up. So that way, you know, they know what the system can and can't do. Finally, we say, hey. We wanted to have certain repository access and we wanted to, you know, only to be able to read the code for now. Now on the for each individual agent, an agent is only as power is as powerful as the sandbox that you can use for the, for where it runs. So in this particular instance, we have, we we give it, you know, a particular set of of, tool sets, whether it be SAM tools, BWA, STAR, and we give it resource limits within that specific sandbox. So all these things will be configurable for your preferred level of automation. Right? So if you got if you want your agent to have full permissions, you're gonna be able to do that. If you don't want it to if you want it to be more locked down, you're gonna be able to configure that as well. The the easiest way for additional, additional, configuration is if you provided a customer a custom Docker file which is where you'll be able to define exactly what, you know, what specific tooling it has access to within in sandbox. This is for folks that are doing, you know, very custom analysis or very specific levels of analysis, you can kind of tell the very specific, packages that your agent will have access to. So for example, if you're doing a lot of single cell work, you'll want to have access to Seurat. If you're doing a lot of work for a taxi, you know, you'll want to have access to defined or a package like that. And so this is where the cuss that custom level of customization comes in. Now the hardest part of these agents, however, is the is understanding exactly what happened. And so within our preview interface, what we wanna do is we wanna show you exactly what an agent has done and exactly which pipelines it has launched. Right? And so the major things here is that we, you know, one is that we have lineage as a first class, as a first class citizen, which, you know, a launch pipeline can, you know, so that way you understand exactly what happened with pipelines that the agent has launched, not just the actions that that it took. And so this is more or less the vision that we are seeing with our agent preview where things are going where, you know, and a lot of the features that we're hoping to bring folks on the call in 26.two. Okay. Cool. I'm going to go ahead and jump back to the presentation. The major takeaway is that Secured Cosciences is your central agent for bioinformatics. Everything that I showed you today, it runs in your infrastructure. It runs on your models with your, with your security, posture. And we where we wanna push things is much more towards, you know, your exact level of comfort with automation and with agents doing tasks within your within your, you know, bioinformatics factory that you are setting up. But where we really, you know, where we really want to kind of enable things is really moving folks into this world where you have certain events that augment context and rules that are stored neatly within, the secure platform that then augment the actions that your agents, your employees, your deterministic actions can take and produce the relevant outputs for your organization. With that being said, we are fielding a ton of folks who, you know, love interest and love kind of exactly where things are going. If you'd want to join, you know, one is the current cosigness product and a lot of the preview see a lot of the preview features that we are working on now. This is the link to, to, to use. So you can either go ahead and scan it with your phone or you can go to sikara.iocosign to register interest. I'm gonna delete Lizzie who's on the in the chat will paste a link, specifically to this form. Okay. Cool. So that's, that's all we have for cosigness. I know we've been talking quite a lot, and then it looks like we have some questions here. Yeah. I'd love to kind of, get get through some of them. Awesome. Thank you so much, Sasha, for the walk through and showing us Secara Cosigness. We actually got some really good questions. So let's go ahead and tackle some of them. So one of the more popular ones, Sasha, how does Secara Coscientists compare to Secara AI? That's a great question. So a the a our our first versions of the product were very, were not were very, active products, meaning that you had to babysit them and there was a lot of work that you had to do in order to connect it with the rest of your systems. What we did is we looked at how our power users were using the system and we saw it was that people, one, had several threads going at the same time and then two, they were using much more of our connected features as much as possible. So what we did is we made the integration with the Secera platform dramatically stronger. We increased the thinking budgets, of our models to be more intentional And we added queuing to our task system so that way a user couldn't work on unlimited number of tasks and sorry, an unlimited number of tasks at the same time. So especially for users that are big fans of the web product, this was a huge unlock for their productivity because they could have multiple tasks running at the same time and multiple kind of agentic loops working on the really hard hard problems that they need, help with. Second question. So a lot of upvotes around. Can you please clarify which LLM provider is powering Cosignton? Is it OpenAI, Claude, or Gemini? And, also, what data privacy residency, and compliancy compliance controls are in place, particularly for sensitive research data in the pharma. landscape? For sure. So, the, the provider current the provider that's doing most of the heavy lifting right now is Anthropic. The our pricing policy is, available on our website. Lizzie, can go ahead and share that. The TLDR for, the non lawyers in the in in the chat is that we don't use your data to to train the, to train our own our own models. But we know that that's not enough for many that that level of, you know, of, security is not enough for many of our pharmacy partners. So when you do buy Secure Secura platform, the enterprise version, and you buy the cosigness product, we actually use your LLM inference endpoints. So whatever agreement that you have with your LLM provider, that's what governs it. And so we don't, you know, we don't, we don't we can't really even see what how you're using those those systems. So you have full control over over what it is, that you're that you're prompting it with. Thanks, Sasha. There's a question around when runs fail. Can we be specific about how to use cosigntest when a Nexo run fails? How can you use it to screen log files and address the root cause analysis? Yeah. That's a great question. So I'm gonna I'm gonna go ahead and just reshare reshare my screen here, just to show you guys how to select that. So within the Secera AI interface, you it's a it's a Secera Cosign's interface. You can click the plus sign here. You can click on can click on runs, which will display all of I'm sorry. I don't have any runs in this workspace. One second. Let me re log in. And we can see, all of the runs here. And then so what I can do now is I can select this one and then say, hey. Please tell me what happened. And it'll go ahead and kind of, diagnose diagnose this specific, this specific run. So that's the that's the best way of doing it. If you're doing it within using the CLI product, the CLI is, trained to look at your dot log files. It will look at the dot nexto dot log. If it notices that it was run with Secura, it will also look at your remote, at the remote logs within the platform to try to diagnose as many of the issues as possible. A good follow-up on that is a question here. How does Coscientists consume underlying compute logs, or which cloud platforms are currently supported? So, if you if you're using the, if you're using the Cloud product, we, we support any platform that you that we onboard. So if you're using the Cloud product to run your runs sorry, to run your next flow, we we support it and we can traverse those we can traverse those logs. If you're an enterprise customer, right now we have current support for AWS, with GCP and Azure coming, coming shortly. Great. Question here. And so have just. to ahead. and sorry. Just to follow-up quickly on that, the number one kind of, source of bugs that we hear about, people telling us anecdotally about cosignancy resolving are infrastructure bugs, not exactly bugs in the code. So a lot of the times and, most of our customers say that the most painful portions of their bioinformatics experience is dealing with, is dealing with AWS and GCP and Azure. That's where our cosigness agent does a really good job of traversing all of those logs and telling you exactly what to which permission to ping IT about in order to unblock you. There's a question here. Do you have any benchmarking data available for COSA test? Yeah. So for cosignists, we do have our benchmarking data here. This is when we benchmark against the latest most powerful version of Cloud Code. As you can see, there's a 20% relative lift on these benchmarks. Now the one thing that if you have to kind of, let folks know is that benchmarks as a whole, and you may see this, you know, with other evals, is that they are run typically on open open code. Right? Meaning that code that is publicly available that model providers and LLMs have likely already trained on. Where we why we see improved performance in the real world is because our token consumption for our harnesses is dramatically lower. So for harder tasks, our context rod is dramatically less, which is why when you're using coscientists, you'll feel you should especially if you're bioinformatics tasks, you should feel more than a 20% lift over vanilla clogged code and things like that. There was a question here, I think, just trying to better understand the mechanics of using the cosign to CLI. So the CLI is running the AI locally question mark, like we load the models and such, or is it just an interface between the local data and a model slash agent that is remote? Yeah. Great question. So it is a interface for your model that is hosted by your pharma provider. Right? The, the the one of the main differences between the way that we, orchestrate versus other, CLIs is that the CLI the harness for us runs even within the CLI is shared with its web counterpart. It's just using your terminal as, as an interface. What this does is a couple of interesting things from a security standpoint that make it much easier for your security teams to onboard because they only have to onboard one kind of one kind of shared back end. So that way, you know, it makes it very, very easy for them to adopt a agent system. So that way they can run cosign to this from the web and from the CLI. There are some questions here around running pipelines on internal infrastructure that's not necessarily on the cloud. So can we still use secure platform and coscientist to examine and debug runs even if we're not using cloud compute, but maybe internal bare metal infrastructure? Totally. It will still work, especially if that bare metal infrastructure is generated in an excel dot log. It just won't be as nice as the Cloud product, that unfortunately, there's nothing we can do about that. But, yeah, it'll totally the the CLI does have access to your, to your terminal. So if the compute is accessible via the terminal interface, it can absolutely be used there. And one more. So beyond a registry, what kinds of extra secure platform, next level context, I. E. Skills, is coscientist automatically provided? Yeah. So there's a ton of proprietary skills that we load into the system, which are meant for pushing a lot of the, a lot of the, a lot of the performance of the harness. The majority of everything that we do is accessible through the MCP. So when you install cosign to this, you are also installing the MCP. So what that does is that it does, it does two things. Is that one is that the MCP is doing a lot of the heavy lifting to make our cosign just amazing, but it's also making your existing your existing agent way more cost conscious and way more productive when you're using our out of the box pre configured modules and things like that. Awesome. Thanks, Sasha. It looks like we are approaching the end, of the webinar. We do have a poll open currently. We'd love if you could answer that. Just let us know what would be most important to you and most useful for you to learn more about coscientists, see more of coscientists. And I think we'll end off, there. But any last words, Sasha, from your end on coscientists? Yeah. Thank you very much everyone for, for attending. We'd love the interest in the product. Right now, we have a full queue of people that are trying to, to onboard. So if you are interested, you know, please, let us know because the sooner, you know, the the sooner you register your, your interest, the sooner we'll be able to, to get to you. For those of you that use the cloud product, I am all over all of the socials, all of our feedback things. If you want things to change, please, please, please, please, please let us know. If there's anything particularly unique about your use case, also please let us know. We try to do our best with our internal evals, but we know that we don't capture all the complexity of real science. And so we're we're trying our best to make sure that these systems are, you know, don't just survive emails, but survive contact with the real world. Alright. Thank you all so much for attending, and for those that asked really great questions and for all the conversations in the chat. We're looking forward to hearing more of your questions and following up on cosign to shortly for those that are interested. So thank you all for your time.