Infrastructure experiment

AgentLab

An experiment in AI development infrastructure: planning, building and review roles, with local inference behind a gateway that makes routing and validation explicit.

Status: Experiment
Type: Infrastructure experiment
Platform: Self-hosted

Current AgentLab gateway architecture: deterministic L0 tasks, L1/L2 local inference through Ollama, schema validation, bounded repair, and metadata-only telemetry. Guard rejection requests escalation without making a cloud call. — AgentLab — gateway architecture

Architecture diagrams, not screenshots of a running system. The workflow roles reflect an infrastructure audit from 3 October 2026.

01 / Gallery

AgentLab visuals

Select an image to see it at full size.

AgentLab workflow roles: Claude planner, Qwen/OpenCode builder, and QA/safety reviewer. Runtime status is based on the 03 October 2026 audit, not a new live check. — Workflow roles — planner, builder and reviewer

More information coming

AgentLab is a work in progress, and this page only states what has been confirmed so far. Still to come:

Screenshots of a running interface (none captured yet)

Questions in the meantime: [email protected].