Workflow-driven AI-agent platform

What lmctl does

lmctl-next runs structured workflows that spawn native AI provider CLIs as cooperating agents. A workflow definition controls sequencing and routing, so the pipeline is the organizing layer.

Provider sessions are treated as disposable caches. Project knowledge lives in durable-memory files, so useful context can survive compaction, restarts, and provider changes.

Tutorials

Install & first run

Install lmctl-next, initialize providers, and confirm the local setup.

First workflow job

Create a project and run the image-qa workflow end to end.

QA suite

Use ai-test chapters to drive repeatable manual checks through workflows.

Manuals / Reference

CLI / API reference

Lookup commands for projects, teams, workflows, jobs, runs, attentions, and issues.

Workflows & archetypes

Understand workflow JSON, compound primitives, and routed outcomes.

Architecture overview

See how the daemon, SQLite state, provider sessions, workflows, and durable-memory fit together.