Everant Studio

Building multiple products solo doesn't scale with one pair of hands, and a lot of that work is genuinely repetitive: research, small fixes, admin, the kind of thing that eats hours without needing real judgement each time.

If the same framework that handles that repetitive load for me could be re-deployed per client with nothing more than a new config file, it stops being internal tooling and becomes something sellable in its own right.

Eight Claude Code agents, each its own macOS user account with a defined role (ops, research, backend, design and more), run against a shared, SQLite-backed task queue with priority routing (a lighter model for scheduling work, a stronger one for real implementation), budget tracking with 80%-alert and 100%-hard-block thresholds, and a notification-cooldown system so it doesn't spam Slack. It doesn't write and ship production code end-to-end yet. What it does today is streamline and take on the small, repetitive tasks that would otherwise eat my own time, and I'm actively working on building it into something that can take on more. A Slack bot is the primary interface for status, task assignment and alerts, backed by a dashboard reachable remotely over Tailscale. It's config-driven by design, which is exactly why it's being built as a product, not just kept as internal tooling.

• Keeping each agent's login details correct even though the system restarts and hands off tasks constantly. Get this wrong and a task fails quietly, with no obvious error to point at, so every agent now reads its own credentials fresh rather than trusting them to still be valid.
• Telling the difference between an agent that's genuinely working and one that's silently stuck, solved by having every stage of a task report its own progress in real time, rather than waiting blindly for a final result that might never come.
• Making sure eight agents working at once don't collide or overwrite each other's work, which needed a proper fix to how tasks are shared between them, not just a quick patch that would break again under load.
• A working demo showing the same framework reskinned across four completely different industries (an estate agency, a financial consultancy, an e-commerce brand and a social media agency), each with its own agents, workflows and brand, proving the config-driven claim rather than just stating it.
• A second demo showing the range extends beyond agent-persona dashboards too: a CRM, an automated chat-support bot, and a data-pipeline and reporting platform, all built to the same production standard.

The interesting problems here were rarely about the agents themselves. They were the unglamorous infrastructure underneath: environment variables, file permissions, and knowing the difference between an agent that's thinking and one that's stuck.

In active daily use, taking on the repetitive research, admin and small implementation tasks across every project on this site, freeing up time for the work that actually needs a person. Being actively developed further, with the goal of reliably taking on more over time.