Reason
Analyze the task, split it into work, and identify dependencies and risks.
Applied AI in Autonomous Systems
Open-source systems research
A provider-agnostic study in autonomous software delivery. The system coordinates 41 agent types across 8 swarms through the Reason, Act, Reflect, Verify cycle.
The control loop
RARV turns a broad goal into a repeated evidence loop, so action is always paired with reflection and verification.
Analyze the task, split it into work, and identify dependencies and risks.
Execute with specialized agents drawn from the right swarm and provider.
Review outputs, inspect evidence, and identify weak assumptions.
Run the gates, compare against acceptance criteria, and prove completion.
Architecture
01
41 specialized agent types organized into 8 coordinated swarms. Each owns a narrow competency and swarms organize around the task.
02
One orchestration layer over Claude Code, Codex CLI, Gemini CLI, Cline, and Aider. Route different work to different providers.
03
Provider-maintained integrations now connect agents to external systems. LokiMCPUniverse is deprecated. Official provider-maintained implementations replaced these MCP servers.
04
OIDC auth, audit logging, sandboxed execution, and Prometheus metrics for regulated environments.
Installation
bun install -g loki-mode loki start ./my-prd.md
npm install -g loki-mode
brew tap asklokesh/tap && brew install loki-mode
docker run --rm asklokesh/loki-mode start prd.md
loki demoRun a demonstration projectloki quick "task"Run a focused one-off taskloki initInitialize a new projectloki self-updateUpgrade through the current package managerLokiMCPUniverse / Deprecated
Official provider-maintained implementations replaced these MCP servers. This catalog is retained as a historical record.