Everything we know about building AI agents, in one place.
Six pillars, every article we've written, organised so you can follow a path instead of wandering. Start with fundamentals if you're new, jump straight to the runtime philosophy or cost economics if you already ship in production.
Start here if you're new to agents or want a precise definition that survives hype. The four ingredients, the loop, the difference with chatbots, and where the wheels usually come off.
How to break a task across a coordinator and specialists, when to spawn workers in parallel, and what it takes to make the dispatch reliable. The architecture under most production setups worth shipping.
Why we stopped writing agents in Python and rebuilt the runtime around a config file. The trade-offs we accepted, the ones that surprised us, and the escape hatch when YAML genuinely runs out.
Real numbers from running coding agents over hundreds of sessions. Where the tokens go, what routing rules actually save money, and the optimisations that look smart on paper but lose in practice.
Ten very different shapes of app the runtime supports, from coding agents to live React sandboxes, slide builders, voice copilots, and scheduled cron reporters. One config file each.
How the encrypted credential vault works on Digitorn. Four scopes, sixteen catalog providers, hash-chained audit logs, and the right way to register a key once and reference it from any agent.
How Digitorn stacks up against the alternatives, including the cases where the alternatives win. We use these stacks daily and the trade-offs are real.
Frontier, fast, open, enterprise. Same YAML shape.
Providers with a green check ship a credential vault entry, you register the key once with digitorn credentials create and reference it from any agent. The others use inline config with env templates today, native catalog entries are on the roadmap.
Battle-tested recipes for the same problems every real agent app hits: retries, circuit breakers, fan-out, routing, human gates, cost caps, context compaction, audit trails.
Install Digitorn and ship your first agent in five minutes.
Open-source runtime, self-hosted by default, your LLM keys stay on your machine. The Hub has the same agents you read about in these articles, ready to install in one command.
