Kanna

Structured autonomy for coding agents.

Kanna lets Claude, Codex, and Copilot work through user-defined workflows: advancing stages, requesting revisions, spawning follow-up tasks, and using the built-in MCP to spawn follow-up tasks and progress through the workflow while each task stays isolated in its own worktree.

Apple Silicon Intel Source

macOS 14+

// WORKSPACE

A desktop control surface for agent workflows.

Tasks, terminals, diffs, files, and pipeline state stay in one place while agents move through the workflow.

Kanna - kanna-web-3

$ kanna-cli stage-complete --task website-refresh --status success

$ kanna mcp create-task "Follow-up copy pass"

review checkpoint selected - approve, revise, or knock back before PR

Diff index.html changed

Files tree explorer ready

Pipeline In progress stage, commit post-action, review checkpoint, and PR handoff.

Worktree Each task gets an isolated branch workspace and terminal context.

Access Desktop and mobile views keep operators in the loop; the built-in MCP gives agents workflow and task controls.

// WHAT IT DOES

Agent autonomy inside explicit workflow boundaries.

Kanna is the coordination layer between coding agents, git worktrees, and the stages your team wants every task to follow.

User-defined pipelines

Define your own agent pipelines. Agents can use the built-in MCP to spawn new tasks or autonomously progress through the workflow, while user-defined human-in-the-loop checkpoints decide where review, approval, or revision is required.

Agent workspaces

Run Claude, Codex, and Copilot inside isolated git worktrees. Kanna automatically provisions per-workspace ports from user config for dev servers and related services, so local apps can launch without colliding.

Definable agents

Add repo-specific agents like merge queue or shipping agents. Desktop and mobile views keep operators in the loop; the built-in MCP gives agents the control surface for stage changes, task creation, and handoffs.

// WORKFLOW AUTONOMY

Agents move through the workflow, not around it.

Configurable stages let agents keep working autonomously through in-progress work, commit post-actions, reviews, and PRs, while named agents handle merge queues, shipping, or repo-specific handoffs.