opencode-context-tree

Pi-style context tree for OpenCode, with the pi-context-tree git-style workflow (/branch, /merge, /crop, /undo) and a DeepSeek-Harness-style trajectory view (timeline lanes, per-step tokens — the model's own where it reports them, estimated ~ elsewhere — timing, inspector) in one screen.

Status: 0.1.0 — first release. Install with "plugin": ["opencode-context-tree"] in both opencode.json and tui.json (see docs/USAGE.md). Branch, jump, labels, filters, search, crop + undo, squash/discard/tournament merge with a $EDITOR gate, the timeline lanes, inspector and consumers views, the gauge, and the headless /ctree commands all work against OpenCode 1.18 and are covered by pty-driven e2e tests (bun run test:e2e). Pressing ⏎ on an earlier message is Pi's fork flow whole — one question with Pi's three answers (no summary · summarize everything below that point · summarize with your own prompt), and the summary covers exactly the turns the move leaves behind. Read DESIGN.md — it contains the research (Pi, pi-context-tree, OpenCode plugin/SDK surface, existing plugins, DSH trajectory), the end-user flows, the combined tree + trajectory mockup, the data model, architecture, edge cases, and the roadmap.

Install

Requires OpenCode 1.18 or newer. One command registers both halves of the plugin (the server plugin in opencode.json, the TUI plugin in tui.json):

opencode plugin opencode-context-tree -g     # for every project (~/.config/opencode)
opencode plugin opencode-context-tree        # for the current project only (.opencode/)

Restart OpenCode. /tree (or ctrl+q) opens the tree; /branch, /merge, /decisions and the headless /ctree commands are available too. The trajectory panels are off by default so the first screen is the plain tree: inside /tree press 1, 2 or 3 for the Input/Model/Tools timeline (0 hides it) and i for the inspector.

If you only see /ctree and no /tree, the TUI half is not registered: run the command above. To register by hand instead, the package name must be listed in both files:

// opencode.json (or ~/.config/opencode/opencode.jsonc)
{ "plugin": ["opencode-context-tree"] }
// tui.json     (or ~/.config/opencode/tui.json)
{ "plugin": ["opencode-context-tree"] }

Upgrading. OpenCode keeps the version it installed and does not re-check @latest on restart. To move to a new release, pin it (this also rewrites both config entries):

opencode plugin opencode-context-tree@0.2.0 -g --force     # or without -g for the project

or delete ~/.cache/opencode/packages/opencode-context-tree@latest and restart. ? in /tree and /ctree status show the version you are running.

Cost of the plugin itself: /tree opens in about 35 ms whether the session has 57 or 467 messages, the crop transform adds under a millisecond to each model request, and the TUI grows by about 37 MB (measured; see docs/USAGE.md).

Options go in either file: [["opencode-context-tree", { "storage": "global", "jumpSummary": "never" }]] (see docs/USAGE.md). To hack on it from a checkout, see "Try it from source" below.

The idea in one screen

/tree is an outline of the whole session tree — Pi's tree — where every message and tool step is one content-forward row and branches are drawn at their fork points:

┌ Context tree · Fix flaky test · trunk                                  ctx ~46k/200k · filling
│ filter: default 24 rows
│ ● user: build yourself a tool that reads the context window…                              ~1.2k
│ ○ assistant: I'll start by inspecting my environment…                                      0.3k
│ ⚙ [bash $ ls -la ~/Documents/] → total 744 …                                              ~2.1k
│ ● user: decompress the session and show the structure                                      ~0.2k
│ ╰⎇ try-redis  ▸ squashed · 9 turns                                                          ~22k
│ ╰⎇ fix-flaky  ▾ open · 6 turns  ← here                                                      ~14k
│ │ ● user: the bun test is flaky, find the race                                             ~0.4k
│ │ ⚙ [bash $ bun test src/foo.test.ts] ⚠                                                     ~4.7k
│ ◆ Decision: try-redis · Outcome: switched to a write-through cache…                         ~0.9k
└ ⏎ go  b branch  m merge  c crop  u undo  s consumers  ? help  q back

Every message and tool call is a row; the gutter draws each branch at the message it was forked from; from anywhere you see the whole tree, your current branch open with ← here and the rest folded. It stays close to Pi so the screen is familiar to anyone coming from it. The DeepSeek Harness trajectory is one keystroke away, not gone: i opens the inspector (per-step payload, result, timing) and 1/2/3 bring in the Input / Model / Tools lane minimap. Sessions made with OpenCode's own /fork are adopted into the tree automatically.

Screenshots

Real OpenCode 1.18 TUI, gemma4 via ollama. /tree is a Pi-style outline of the whole session tree: one row per message or tool step, thinking folded into its assistant row, branches drawn at their fork points and folded to their ⎇ header until opened, the context bar and band in the header, and a footer that says what ⏎ will do for the selected row:

From inside a branch, ← here marks your step and the trunk rows past the fork point are dimmed under ── not in this branch's context ──: they are not sent to the model.

/ filters as you type, highlights matches and counts rows; n/N step through them:

The DeepSeek-Harness trajectory is one keystroke away: 1/2/3 bring in the event-strip lanes (one pill per event, width by duration, red for a failed tool call) and i the inspector:

s shows what is filling the context, by share of the tree and of the model window, expandable into entries you can crop in place; D renders the decision records; ? opens the help pane:

The gauge on the prompt line, here on a caching provider: the context of the next prompt (the same figure OpenCode's own sidebar shows), its band, and how much of that prompt the provider served from its cache. The bar's dim cells are the cached part; 0% cached right after a crop, merge or fork means the cache was reset.

Try it from source

bun install && bun run build
# opencode.json  →  "plugin": ["/abs/path/opencode-tree/dist/server.js"]
# tui.json       →  "plugin": ["/abs/path/opencode-tree/dist/tui.js"]

Keys inside /tree (vim-aligned): j k ctrl+d ctrl+u gg G move · [ ] hop between branches · h l Tab fold/unfold · ⏎ go (the footer says what it will do for the selected row) · b branch · m merge · c crop mode (space mark, a auto, t result⇄turn, ⏎ apply) · u undo (x too) · / live search, n N next/prev · f filter picker · i inspector · 1 2 3 lanes, 0 off · s what's filling the context · D decisions · L label · y copy · ? help · q.

Commands

Command	What it does
/tree (Ctrl+Q)	open the combined tree + trajectory view
/branch <name> [model]	fork here into a named branch, optionally on a cheaper model
/merge [--pick | --no-llm | --discard | --tournament]	close the branch: Squash drafts a ◆ decision record you confirm, Squash without LLM lets you write it, Discard lands nothing, Tournament keeps one of several siblings. Your transcript is never rewritten; the record is appended to the trunk as a normal message
/crop [--top | --auto …]	stub fat tool results or drop whole turns from what the model sees; append-only, reversible
/undo	revert the last branch / merge / crop
/decisions [--export]	list or export decision records

How it maps onto OpenCode

• a branch is an OpenCode session created with session.fork; the plugin remembers (parent, anchor) in an append-only journal and mirrors it into session.metadata;
• crop is applied per request in experimental.chat.messages.transform, so the transcript keeps the originals and the model sees stubs;
• merge writes the confirmed record with session.prompt({ noReply: true });
• the UI is a TUI plugin (@opencode-ai/plugin/tui): a route, two slots (gauge, sidebar card), dialogs and a keymap layer.