specialized agents
L-O-D-U-R-R / Local Orchestration Daemon for Unsupervised Reasoning & Repair
Lodurr
A local-first runtime for governed AI workflows. Lodurr brings message routing, execution control, multi-agent memory, tool orchestration, evaluation loops, and human oversight into one coherent product system.
memory strata
evolution loops
Hermes Mesh
Agent Harness
Memory Mycelium
Tool-use Cortex
Project Kernel
Eval Reflex
Human Gate
Open-ended Adaptation
Hermes Mesh
Agent Harness
Memory Mycelium
MODULES / agent runtime capabilities
Composable modules for building reliable agent systems.
Lodurr organizes agent work into clear operating layers. Each module owns a specific responsibility: route intent, govern execution, preserve useful memory, evaluate outcomes, and keep humans in control.
KERNEL / agentic architecture
Lodurr is not a model layer. It is the runtime for governed agent work.
A single agent can quickly become a large prompt with unclear responsibility. Lodurr separates planning, action, memory, evaluation, and policy so agent workflows become easier to operate, inspect, and improve.
Human Gate
Intent, approval, boundaries
Hermes Message Bus
Task routing and context compression
Agent Colony
Planning, research, build, review, release
Harness Runtime
Tool permissions, outputs, checkpoints
Memory Mycelium
Operational memory and reusable skills
INTRO / product manual
A concise product surface, with the full manual in GitBook.
Lodurr is a local-first Agent Runtime for teams that need structured orchestration, durable memory, evaluation loops, and controlled execution. This page presents the product. The GitBook contains architecture, operating principles, deployment guidance, and developer workflows.
Read the complete Lodurr manual: product model, architecture, Hermes and Harness principles, mathematical reasoning, local operation, Docker, GitHub Pages, CI/CD, security, and release strategy.
Open GitBookOPERATIONS / learning loop
Every workflow leaves evidence for the next improvement cycle.
Lodurr turns operational activity into reviewable signals. Plans, actions, outcomes, memory updates, and human decisions become part of a continuous reliability loop.
Product goals are converted into structured work packets.
Planning, build, evaluation, and release lanes receive clear responsibilities.
Repeated decisions become reusable operating knowledge.
Evaluation signals define the next reliability upgrade.
Optimization queued: operating boundaries, memory policy, and evaluation criteria are aligned.