Autogent Lab // knowledge base
Lab collects the technical notes, product lessons, reference architectures, and deployment patterns we use when building agent runtimes and delivering them into real organizations.
01 / Tracks
How agent programs execute, recover, coordinate, and expose state.
The infrastructure work that determines latency, margin, and scale.
Boundaries for agents that read files, browse the web, and run code.
What we learn from shipping real agent products and enterprise POCs.
02 / Published notes
Treat the KV cache like a database — paged, evictable, sharable across agents — and the rest of the runtime falls into place.
Temporal is not your agent runtime. LangGraph is not your workflow engine. Naming the boundary makes both better.
Why Firecracker, gVisor, and a plain container are not interchangeable when your agent gets to run shell commands.
03 / Reference architectures
Browser sandbox, source capture, citation graph, reviewer queue, and trace replay for long research tasks.
Private retrieval, structured extraction, reviewer checkpoints, policy constraints, and immutable audit logs.
Model gateway, tool broker, microVM pool, workflow engine, observability plane, and operator console.
Session memory, eval harness, growth instrumentation, cost scheduler, and human-in-the-loop recovery.
04 / Lab queue
