One Planka instance, two surfaces. plnk is for automation; plnk-tui is for an interactive human session. Each has its own credentials — reach for whichever matches the task.
Scriptable CLI for automation, CI/CD, and AI workflows. Strict grammar, typed exit codes, structured JSON / markdown output, machine-readable help.
plnk card find --title "bug" --board $BOARD --output json
Live tree explorer for projects, boards, lists, and cards. Single-board websocket subscription for real-time updates. Edit titles inline, descriptions in $EDITOR.
plnk-tui # prompts for server, username, password on first run
Released 2026-05-02.
Breaking change. The TUI no longer shares plnk's config or environment variables. Launch plnk-tui with no flags and it prompts for server, username, and password on first run, then offers to save the non-secret server + username to ~/.config/plnk-tui/config.toml. Passwords are never persisted. The TUI's environment variables were renamed from PLANKA_* to PLNK_TUI_*; plnk's own PLANKA_SERVER / PLANKA_TOKEN are unchanged. Rationale: plnk is automation/AI/API-token oriented, while plnk-tui is for an interactive human session — sharing one credential store for two very different access patterns was a footgun.
Notes on building plnk, plnk-tui, and working with AI agents in the terminal.
Why I built a TUI companion to plnk — a bridge between the Planka web UI and the agent in my terminal.
Why I came back to Planka and built a thin, strict, hierarchy-aware CLI for AI agents.
Prebuilt binaries for macOS, Linux, and Windows ship on every release, or build from source with cargo.
Tested against self-hosted Planka. Each release also publishes per-platform tarballs/zips with SHA-256 checksums on the releases page.
Planka already does kanban well. plnk wraps it in a tight, typed surface agents and scripts can drive.
project → board → list → card. Every find demands a scope. No global flat queries, no surprising results.
table for humans, json for scripts and agents, markdown for reports. JSON is a strict projection of the serde model.
Scripts branch on 0 success, 2 validation, 3 auth, 4 not-found, 5 server. No stderr parsing.
plnk <cmd> --help --output json emits a schema agents can bind to before ever running a command.
One board live at a time. Edits from the browser appear in the terminal in near real time. No polling, no drift.
plnk-core is usable on its own. Build your own Planka integration in Rust without depending on the CLI.
No account service to sign up for. No config files to hand-edit. You already have a Planka instance — plnk just talks to it.
Grab plnk and optionally plnk-tui with cargo install.
Run plnk init for the CLI — prompts walk you through server URL, API token, and optional HTTP tuning. plnk-tui has its own first-run prompt; just launch it.
I use a self-hosted Planka for everything — technical and non-technical.
What I care about is the artifact. By the time a card hits Done, it contains what I did, the decisions I made, and why.
Many tools in this space are building AI task managers — AI sub-agent drivers, new kanban boards, TODO.md skills, bolting MCPs onto various products. For me, Planka already solves the fundamentals: UI, hierarchy, self-hosting, and a working API. Replacing that felt unnecessary.
I was already driving agent workflows through the REST API. It worked, but it was inefficient — lots of curling, constructing large JSON payloads, and chaining calls for simple operations like finding a card by name. Hard to read, token-heavy, and noisy in the agent's context.
I didn't go the MCP route either. Others are doing it, it inflates context, and I've found it slow.
So plnk doesn't replace Planka — it wraps it in a strict, predictable interface. It reduces token usage and keeps the agent's context lean: the surface is learned on demand through plnk <cmd> --help --output json, not preloaded at session start.
It exposes Planka as a disciplined CLI and SDK — tight grammar, typed exit codes, structured output, strict hierarchy, machine-readable help. Give an agent a rigid surface, and it stays on the rails.
plnk-tui started as an experiment — a live terminal explorer for the same hierarchy, with its own human-oriented login flow distinct from plnk's automation/API-token config. It's fit into my terminal-driven workflow better than I expected, so I'll keep expanding it.