Overview
Looping Louie
Looping Louie is a platform for running review-driven software development workflows with coding agents. It coordinates reusable instructions, Activities, and Pipelines with controlled execution in configured Git repositories.
The platform keeps durable decisions and execution history separate from local repository access. This makes each run observable and allows workers to execute only work that is compatible with their configured environment.
Platform components
The API is the control plane. It manages resources, selects models and Harnesses, coordinates execution checkpoints, and retains durable state.
The dashboard provides a visual interface for configuration and operational visibility.
The CLI brings platform workflows to a developer’s terminal. It manages resources, starts work, and surfaces execution results through the API without owning scheduling or durable run state.
The runtime is the execution worker. It maps Projects to configured local Git checkouts, claims compatible work from the API, and performs approved local operations.
Design a Pipeline
Users compose work from reusable Activities and Pipelines. Activities define the work, its instructions, model, and Harness; a Pipeline arranges them into a review-driven workflow.
Trigger an execution
When a Pipeline runs, the API freezes its effective model, instructions, and Harness selection into durable Activity-run snapshots.
1. Wait for the execution path
The API selects the next eligible Activity. A compatible runtime claims repository work; human Activities wait for a decision from a person.
2. Perform the selected work
The selected Harness determines how an executable Activity performs its local work.
The API-directed louie Harness applies an API-approved operation plan in its
configured checkout.
The codex_cli and copilot_cli Harnesses receive API-scheduled local agent
turns. The runtime invokes the selected CLI with the run’s frozen instructions
and reports the outcome. The API owns the loop roles and decides which turn
follows.
Human approval and quiz Activities do not execute in a runtime checkout. A
person submits their decision through the Activity-run API.
3. Review and advance
The API evaluates each reported result and either schedules more work, records a review decision, retries the applicable loop, or marks the Activity terminal. The API remains the source of truth for workflow state.
4. Complete the run
For commit-enabled work, the API authorizes commit handoff only after the run succeeds. The runtime applies its local Git policy, commits only when permitted, and reports the outcome. The API records the final Activity and Pipeline state for later inspection.
The API owns immutable run configuration, workflow decisions, review outcomes, and durable execution history. The runtime owns local checkout access, local CLI execution, credentials, and Git policy.