Your plan is a star-map.
Your agents fly it.

Chart a map of the work, then drive it to completion. Pick a ticket off the frontier, pick an agent — the session opens with full context already in the buffer.

open source no account no telemetry your agents, your machine
The chartr cockpit: sidebar of spaces, a Claude Code session, and a wayfinder star-map with tickets and dependencies
Spaces on the left. Agents in the middle. The map on the right.

Visible plan, actionable tasks.

Work with an agent on your plan, written as markdown. chartr renders the star-map automatically. Get a bird's-eye view of the frontier of your work.

hover a star. click a frontier ticket to spawn.
  1. 01

    chart — the plan draws itself

    Run the wayfinder skill from the sidebar. It interviews you, writes the map, and the stars appear as the files land.

  2. 02

    spawn — the frontier is the runnable work

    Every ticket whose blockers are answered is on the frontier. Click one, pick a role and an agent — the session opens with full context already submitted.

  3. 03

    resolve — answers unblock the map

    A ticket resolves when its answer lands in the file. The star turns; whatever it blocked joins the frontier.

Never brief an agent again.

Pick a ticket and chartr writes the briefing for you — the plan it belongs to, the question it asks, every decision it already rests on. The session opens with all of it loaded, so the first thing you type is a follow-up rather than a wall of background.

gathered from your repo
  • the mapthe plan this ticket belongs to
  • the ticketthe one question it has to answer
  • the answers behind iteach decision it depends on, verbatim
  • the rolegrill, research, prototype, implement
one briefbuilt at spawn
the session opens
Already up to speed.

It knows where the work sits, what it is here to settle, and what has already been decided — before you type a word.

● no pasting · no re-explaining · no catching up

rebuilt from your files on every spawn · no hidden memory to drift · the map stays the source of truth

Your agents, not chartr's.

chartr ships no models and no agent CLIs. It drives the ones you already pay for.

claude
codex
opencode
kimi
grok
pi
+ yours

aOne seam, every CLI

The adapter models a single thing — how a binary takes its opening line — so anything on PATH registers. Six are detected for you.

bRight brain for the job

Grill a spec with one model, implement the ticket with another. The agent is your call, per spawn.

cNothing your agent can't do

Sessions open in each agent's own TUI — not a rebuilt chat view. Real processes, real PTYs, your keybindings, your themes.

Fully local and hackable. No lock-in.

chartr runs on your machine and writes plain files into your repo — the plan, the method, and the history. Edit any of them by hand to fit how you work. Nothing you own points back at chartr, so it comes off whenever you want and takes nothing with it.

your machine
chartr detach anytime
yours — plain files on your disk
.plan/maps/
the plan

One markdown file per ticket. Read, edit or grep it with no chartr in sight.

.chartr/skills/
the method

Plain SKILL.md files. Your copy shadows the shipped one, per repo or per machine.

.git/
the history

Claim and answer commits, written in ordinary git. Never an amend, a rebase or a push.

Built on a proven workflow.

The maps chartr drives are wayfinder maps — plan by interview, break the work into tickets with blocking edges, resolve it one question at a time. The method comes from mattpocock/skills; chartr ships a fork as its skill library and gives the workflow a cockpit.

Chart the work. Fly the frontier.
Keep the receipts.

rengwu/chartr on github