# Chartr **Chartr is an agent multiplexer with a map of the work: your plan renders as a star-map, and you spawn agents by clicking tickets on its frontier.** Go backend, Svelte frontend, MIT, desktop app for macOS / Linux / Windows. Ships no models and no agent CLIs — it drives the ones you already pay for. ## The idea The plan lives in your repo as plain Markdown, one file per ticket, under `.plan/maps/`. Tickets have blocking edges. Anything with no unresolved blockers sits on the **frontier**, and the frontier is exactly the set of work you can actually start right now. Three verbs: 1. **Chart** — a bundled skill interviews you, writes the map, and the stars appear as files land. 2. **Spawn** — click a frontier ticket. A session opens with the brief already loaded. 3. **Resolve** — a ticket resolves when its answer lands in the file. Whatever it blocked joins the frontier. ## The part that actually earns its keep **You never brief an agent again.** On spawn, chartr assembles the brief from your files: the map, the plan the ticket belongs to, the one question the ticket has to answer, every prior decision it depends on *verbatim*, and the chosen skill (grill, research, prototype, implement, …). The session opens already up to speed. Rebuilt from files on every spawn, so there is no drift — the map stays the source of truth rather than a stale summary of it. That's the real problem it targets. Context re-establishment is the tax you pay on every new agent session, and most tooling attacks it with longer context windows or memory files. Chartr attacks it by making the plan structured enough to generate a brief mechanically. ## Multiplexer side Register any CLI agent on your PATH — Claude, Codex, [[OpenCode]], Kimi, Grok, Pi, whatever — with your own arguments and environment variables. Multiple entries allowed, so you can grill a spec with one model and implement the ticket with another. Per spawn. Sessions open in each agent's own TUI on a real PTY, not a rebuilt chat view. Your keybindings, your themes. Without maps it degrades gracefully into an ordinary terminal multiplexer. ## No lock-in, deliberately - `.plan/maps/` — the plan, plain Markdown, greppable with chartr nowhere in sight - `.chartr/skills/` — plain `SKILL.md` files; your copy shadows the shipped one, per repo or per machine - `.git/` — claim and answer commits in ordinary git. Never an amend, never a rebase, never a push Their line on it: there is no such thing as ejecting from chartr, you just stop using it. Runs on `127.0.0.1:8787`, no account, no telemetry, no analytics. ## Lineage The maps are **wayfinder maps** — plan by interview, break work into tickets with blocking edges, resolve one question at a time. That method comes from `mattpocock/skills`; chartr gives the workflow a cockpit. [[Herdr]] is credited as the multiplexer that inspired it, the difference being that herdr lives in your terminal and chartr in a window. Worth trying alongside the `/wayfinder` skill rather than instead of it — chartr is the GUI over a method that already works as plain files. ## Related - [[Herdr]] — the terminal-native predecessor - [[AI Agents]] - [[AI Agent Skills]] - [[Claude Code]] - [[Codex CLI]] - [[OpenCode]] - [[Kimi CLI]] - [[Agentic Engineering]] - [[Go]] - [[MIT License]] ## References - Website: https://chartr.dev - Source code: https://github.com/rengwu/chartr - Announcement (X): https://x.com/johnrengwu/status/2082063318781718678 - The wayfinder method it drives: https://github.com/mattpocock/skills