How Hyperspell put production on autopilot with Firetiger

Hyperspell, the memory and context layer for AI agents, uses Firetiger's agents to automate production monitoring and fixes, catching Kubernetes crash loops and cleaning up Temporal workflows. Platform engineer Charlie Schluting enabled Firetiger's Autofix to trigger Replicas coding agents that one-shot PRs, with a separate agent verifying and auto-merging fixes, and Change Monitors to proactively detect regressions from new releases.

How Hyperspell put production on autopilot with Firetiger I sat down with Charlie Schluting , the first founding platform engineer at Hyperspell https://hyperspell.com/?ref=blog.firetiger.com , to talk about where Firetiger https://www.firetiger.com/?ref=blog.firetiger.com fits in the way his team ships code to production. Hyperspell is the memory and context layer for AI agents: a "company brain" that connects tools like Slack, Gmail, Notion, and Google Drive, synthesizes them into a live, permission-aware context graph, and serves it to any agent through a single API. Hyperspell runs entirely on AWS, built on Temporal Cloud for durable workflow orchestration, Amazon Bedrock for inference, SageMaker, and everything running in Kubernetes across three regions: US, EU, and Singapore, and Charlie's platform team owns all of it A modern take on system observability Moving to raw Kubernetes gives you control, but also a much bigger surface to watch. The modern way of building infrastructure makes it surprisingly easy to lose track of what's actually happening in production. "It's very easy to assume that something you did is working when using terraform and coding agents." Terraform applies cleanly, your coding agent reports success, and you move on, but it doesn’t mean production systems are healthy. We’re using coding and review agents during development, but we’re still on the hook when we ship code to production. “We had Firetiger before we had Grafana alerts,” says Charlie. Firetiger agents help close the monitoring gap in production, including the failures nobody had thought to instrument yet Firetiger caught and provided complete issue reports for Kubernetes crash loops of pods for which there were no existing monitors. From auto-fix to auto-merge "Firetiger has been really useful in cleaning up our Temporal workflows." Temporal acts as Hyperspell’s orchestration layer, it’s used extensively for mission-critical workflows, so production issues in this system can have a significant impact on the product, setting the bar on correctness and reliability really high. To harden their Temporal-based system as it scaled, Charlie used a combination of: - the Temporal template from Firetiger’s agent catalog - a Firetiger connection with their Replicas account to spawn coding agent - Firetiger’s Autofix feature With Autofix enabled, Firetiger automatically triggers a Replicas coding agent for every actionable issue. The context of the issue is passed to the coding agent, including the full root-cause analysis; in most cases, it allows the agent to one-shot a PR to fix a production issue. "The context provided by the Firetiger issue is really good. We had some really high-quality PRs come out of there." To close the loop, Charlie configured a separate Replicas agent that runs continuously and verifies the PRs opened in response to Firetiger issues. The agent tests the change, and if it can verify the fix, it automatically merges the PR. Those tactical production fixes are later reviewed by humans once the dust settles. By combining those agents, Charlie was able to automate fixes for edge cases that otherwise never get prioritized, to drastically improve the reliability baseline of their system. From reactive to proactive monitors While Hyperspell started with an agent to improve reliability of their Temporal workflows, the causes for production issues quickly shifted from being caused by pre-existing bugs to being mostly caused by changes that were newly introduced. In response to this shift, Charlie enabled Firetiger Change Monitors , another flavor of agent designed to focus on monitoring changes as they are released to production, verifying the intended effects and looking for potential regressions, proactively. "It does the thing that I would never take the time to do — I would never be as thorough. I'm too focused on the change I'm making, but the change monitor is a lot more broad." Over the course of four months, the number of production issues identified by change monitors grew f rom none to more than half of the total volume, while multiplying by 5 the number of PRs merged Change velocity is up, and issues are getting caught the moment they are introduced rather than days or weeks later 🫳🎤 A peek at the horizon Something Charlie kept coming back to is how much of this a small team simply cannot do by hand. The infrastructure surface keeps growing, coding agents keep multiplying the number of changes in flight, and the traditional answer of provisioning more dashboards and more alerts doesn't scale with modern engineering teams. The loop Hyperspell is settling into is the one we're building toward everywhere: coding agents write the change, Firetiger watches it in production, and when a regression occurs, the issue is reported with enough context built over time from the change history, that the fix can be fully automated.