When 2025 started, LoopLlama could do one thing well: run a crew of agents in sequence over a single API call. By the end of the year it had grown into something a team can actually build a product on. Here's the shape of what changed.
From a runtime to a platform#
We spent the first half of the year on the unglamorous foundations — a durable run data model, streaming progress events, typed SDKs, and the security work behind our SOC 2 effort. The second half was about reach: connectors to real systems, bring-your-own tools over OpenAPI, webhooks, and per-step model routing.
- Typed SDKs for TypeScript, Python, and Go.
- Webhooks for completion and step events, alongside streaming and polling.
- Bring-your-own tools via OpenAPI, plus first-party Gmail and Salesforce connectors.
- Per-step model routing, so cheap work runs on cheap models.
The throughline#
Every one of those features came from the same belief we started with: that the hard part of agents isn't the model call, it's everything around it — state, tools, observability, cost, and trust. 2025 was the year we turned that belief into a platform. We think 2026 is the year teams stop wiring agents by hand entirely.