OpenCode TUI plugins that show quota usage, reset countdowns, rate-limit status, compact homepage summaries, MCP server health, active-session context and spend, LSP status, synchronized session TODOs, complete session-tree token totals, direct-child SubAgent activity, and /tokens_* reports for Z.AI (GLM), OpenAI (ChatGPT Plus/Pro), and OpenCode Go.

Requires OpenCode 1.18.1 or newer, Git, and Node.js with npm.

git clone https://github.com/aamkye/opencode-tools.git
cd opencode-tools
npm ci
npm run deploy:global

Fully restart OpenCode after deployment.

• 5H token quota — remaining %, live countdown to next reset, and absolute token counts (used / total) when the plan exposes them (Max/Pro).
• 7D weekly limit — same bar + countdown; shows "Unlimited (Legacy)" when the plan has no weekly cap.
• Peak/off-peak indicator — Peak (14:00–18:00 SGT, 3x usage) vs Off-Peak.
• Limited indicator — shows when the 5H quota is exhausted.
• Heuristic fallback — if the API is unreachable, scans the session's message parts for a reset time and falls back to a clock-based estimate.

• API-reported quota windows — primary and optional secondary windows use compact labels derived from their API-reported duration, such as 5H, 7D, or 1M, and show remaining percentage plus reset countdown.
• Plan type — Plus / Pro / Pro Lite / Team.
• Limited indicator — shows when rate limit is reached.

• Subscription windows — exact remaining usage for rolling 5H, weekly 7D, and subscription month 1M windows.
• Shared refresh behavior — uses the configured polling interval, one-second countdowns, reset-boundary refresh, and a ten-minute stale horizon without exhausted backoff.

• Reactive server health — shows OpenCode's synchronized MCP server list in source order without polling.
• Native status roles — connected, disabled, failed, authentication, and client-registration states use compact labels and status-colored bullets.
• Session-local collapse state — resets to the configured default whenever the active session changes. The collapsed header shows a success/warning/error health rollup (connected, disabled, and error-state counts) that stays compact as a colored 0/0/0 summary when the server list is empty.

• Reactive active-session metrics: updates context and spend values from synchronized session and message state without polling.
• Newest positive assistant token selection: sums finite detailed input, output, reasoning, cache.read, and cache.write buckets and uses the newest assistant message whose sum is positive.
• Cumulative finite assistant spend: sums finite assistant-message costs for the active session and ignores missing or non-finite costs.
• Unavailable values: shows Tokens -, Used -, and Spent $0.00 when the host has not supplied usable context data. When consumed tokens are known but the model context limit is unavailable, the panel preserves the known Tokens value and accumulated Spent, while Limit, Used, and the collapsed summary remain -.
• Session-local collapse state: resets to the configured default whenever the active session changes.

• Reactive server list — shows OpenCode's synchronized LSP IDs in source order without polling.
• Operating directory — each row shows the basename of the server's root (the directory where it operates), right-aligned in the muted label color. Servers with an empty root show the id alone.
• Status-colored bullets — successful servers use the success color, failed servers use the error color, and unknown statuses remain in the list with a muted bullet.
• Session-local collapse state — resets to the configured default whenever the active session changes and shows LSPs will activate as files are read when the expanded list is empty.

• Synchronized session scope — shows the active session's TODO records in source order without polling.
• Status markers — uses [✓] for completed, [•] for in-progress, [ ] for pending or unknown, and [-] for cancelled records.
• Aligned wrapped rows — reserves four cells for each marker so continuation lines align beneath the content column. An empty list shows No TODOs for this session.
• Session-local collapse state — resets to the configured default whenever the active session changes and summarizes done/working/todo counts (cancelled excluded) in the collapsed header.

• Assistant-only full-tree totals: aggregates assistant messages across the selected root session and its complete descendant tree.
• All detailed token buckets: reports input, output, reasoning, cache read, and cache write. Total is input + output + reasoning + cache read + cache write; cache hit ratio is cache read / (input + cache write).
• Compact absolute counts: uses K/M/B suffixes with up to two decimal places and trimmed zeroes. The collapsed summary shows only the aggregate total.
• Event-driven refresh: uses a 200 ms event debounce and 2, 4, and 8 second retries. It does not poll or calculate cost.
• Stale recovery: retains the last successful snapshot as stale and recovers it to ready after a successful refresh.
• Memory-only data: Snapshots and collapse choices are memory-only. Selecting another session resets the panel to its configured or built-in default.

• Direct-child monitoring: monitors direct child sessions of the selected parent, ordered newest first. The newest five stay in the primary group and older children appear under Rest.
• Status precedence: failed takes precedence over running and successful; busy or retry is running; idle or a completed assistant message is successful; otherwise the child remains running.
• Duration and details: each row shows a live duration for running children and a terminal duration for successful or failed children. Expanding a row shows its agent, status, time, model, and Open Session action. Compact durations and expanded time values use the child's status color. Compact titles are grapheme-safe and end-truncated beside a fixed seven-cell, right-aligned duration box with a two-cell structural margin. Expanded titles wrap in full without a duration reservation. The Rest disclosure and title are muted, and its divider is two muted three-dash segments separated by flexible space.
• Session-local interaction state: panel collapse, Rest collapse, and an expanded child reset whenever the active session changes. Retained failure evidence remains persistent per parent session.
• Stale and empty behavior: a failed refresh after ready data retains the complete entry body and marks the panel stale. Loading, unavailable, and unselected states emit no panel output; a ready parent without direct children shows No subagents.

• Homepage summary — each provider plugin also registers a compact homepage line, such as Z.AI: Max; 93%/84% or OpenAI: Pro Lite; 96%/84%.
• /tokens_* commands — server plugin providing token usage and cost reports: /tokens_today, /tokens_daily, /tokens_weekly, /tokens_monthly, /tokens_all, /tokens_session, /tokens_session_all, /tokens_between. Reads from opencode.db with full models.dev pricing resolution.
• Color-coded bars — green above 30% remaining, amber at ≤30%, red at ≤10% remaining.
• Provider names, plan types, and bar labels use the theme foreground colour; only the bar fills and percentages are colour-coded.
• Smart polling (Z.AI and OpenAI) — checks the quota API every 10s by default, backing off to 5min when the primary window is exhausted.
• Expandable — click the header to show weekly / tool / absolute details.
• Stale handling — keeps showing the last known data through transient fetch failures and marks stale in the right-aligned provider header.

The plugins are built and loaded only from local files. This package is not published to npm, and OpenCode is never configured with an npm package spec. OpenCode 1.18.1 or newer is required for the standalone TUI plugin and synchronized MCP, Context, LSP, TODO, session-tree, and SubAgent state APIs.

Native TUI options can be supplied with the local plugin entry:

{
  "$schema": "https://opencode.ai/tui.json",
  "plugin": [
    "./opencode-tools-home.js",
    "./opencode-tools-token-report.js",
    "./opencode-tools-context.js",
    "./opencode-tools-ses-tokens.js",
    "./opencode-tools-subagent.js",
    [
      "./opencode-tools-quota.js",
      {
        "quota": {
          "refreshIntervalSeconds": 10,
          "progressColors": {
            "enabled": true,
            "errorBelow": 10,
            "warningBelow": 30
          },
          "percentageMode": "remaining",
          "hideInactive": false,
          "openai": { "hideInactive": false },
          "zai": { "hideTools": false, "hideInactive": false },
          "opencodego": {
            "workspaceId": "wrk_TESTWORKSPACE",
            "workspaceToken": "TOKEN_TEST_ONLY_DO_NOT_USE",
            "hideInactive": false
          },
          "otherProviders": { "sortDirection": "desc" }
        }
      }
    ],
    "./opencode-tools-mcp.js",
    "./opencode-tools-lsp.js",
    "./opencode-tools-todo.js"
  ],
  "plugin_enabled": {
    "internal:sidebar-context": false,
    "internal:sidebar-mcp": false,
    "internal:sidebar-lsp": false,
    "internal:sidebar-todo": false
  }
}

Context ships as a separate opt-in artifact. Enable it by adding ./opencode-tools-context.js to the plugin array.

The entries must remain standalone and in manifest order. Quota accepts the quota options object; each sidebar panel (Context, SesTokens, SubAgent, MCP, LSP, TODO) optionally accepts an options object with defaultState and chip. Home and token-report use string entries.

Any sidebar panel can be configured with defaultState. Whenever the active session ID changes, including returning to a previously opened session, every panel resets to that configured state. A missing or invalid value uses the built-in "expanded" default. Header and secondary-section toggles remain local to the currently selected session and are not persisted.

"semi-collapsed" starts the panel expanded with the secondary section (Rest / Other Providers) collapsed; it is ignored for panels without a secondary section.

["./opencode-tools-context.js", { "defaultState": "collapsed" }]

MCP, Context, LSP, and TODO have no built-in panel override to disable. SesTokens has no built-in panel override. The SubAgent has no built-in panel override. The other external panels do not deactivate their built-in counterparts. Users must disable internal:sidebar-mcp, internal:sidebar-lsp, and internal:sidebar-todo themselves, as shown by plugin_enabled, to avoid duplicate panels.

quota.opencodego.workspaceId identifies the OpenCode Go workspace. quota.opencodego.workspaceToken authenticates the console request; workspaceToken is the plaintext auth cookie value. Keep both values only in local .opencode/tui.json: they must not be committed or shared, and you must rotate the console session when it expires, is revoked, or is exposed.

The provider sends these workspace credentials only to the fixed https://opencode.ai origin; they do not replace the OpenCode-managed inference API key. The sidebar reports exact remaining usage for rolling 5H, weekly 7D, and subscription month 1M windows. OpenCode Go reads the undocumented Solid hydration contract from the authenticated page and fails closed if that contract changes. It does not scrape visible text, save page HTML, or estimate quota from local cost.

OpenCode Go uses the shared default/custom polling interval, one-second countdowns, reset-boundary refresh, and a ten-minute stale horizon without exhausted backoff.

Defaults are quota.refreshIntervalSeconds: 10, quota.progressColors.enabled: true, quota.progressColors.errorBelow: 10, quota.progressColors.warningBelow: 30, quota.percentageMode: "remaining", quota.hideInactive: false, quota.zai.hideTools: false, and quota.otherProviders.sortDirection: "desc". Provider hideInactive overrides resolve as providerOverride ?? quota.hideInactive ?? false. Inactive controls affect only configured providers that are not selected; the selected provider remains visible. Set quota.zai.hideTools to true to remove every Z.AI tool-limit row and its quantities.

Polling defaults to 10 seconds when its value is invalid or non-positive. Color thresholds are clamped to 0-100, and errorBelow cannot exceed warningBelow. Set quota.progressColors.enabled to false to disable semantic bar and percentage colors.

The root-level refreshIntervalSeconds and progressColors paths are ignored. Move them to quota.refreshIntervalSeconds and quota.progressColors respectively. The legacy root-level otherProviders object is also ignored: move otherProviders.percentageMode to quota.percentageMode and otherProviders.sortDirection to quota.otherProviders.sortDirection.

Every configurable option accepted by the plugins and host-level files is listed below. Options are supplied through the tuple form ["./plugin.js", { ... }] in tui.json (or opencode.json). Unknown keys in an options object are ignored.

Per-plugin options

defaultState resets every sidebar panel when the active session ID changes. "semi-collapsed" starts the panel expanded with the secondary section (Rest or Other Providers) collapsed; it is ignored for panels without a secondary section. Missing or unrecognized values fall back to "expanded".

Input chips (chip option)

The seven sidebar-panel plugins (Quota, Context, MCP, LSP, TODO, SesTokens, SubAgent) also render a compact status chip through session_prompt_right, on the right of the in-session prompt's agent/model row. Each chip reuses its panel's collapsed summary and semantic status colors — e.g. Q 46%, Ctx 64%, MCP 4/0/0, LSP 2, TODO 4/3/2, Tok 29.11M, Sub 7/1/3. A plugin renders no chip when its panel has no data. Home and token-report do not render chips. Chips are display-only and do not affect the sidebar panels.

chip: "disabled" suppresses only that plugin's input chip. Missing or unrecognized values fall back to "enabled".

For example, disable only the LSP chip while keeping its sidebar panel enabled:

["./opencode-tools-lsp.js", { "defaultState": "collapsed", "chip": "disabled" }]

Quota plugin options (quota object)

When progressColors.errorBelow exceeds progressColors.warningBelow, both revert to their defaults (10 and 30).

Host-level configuration

Built-in panel overrides:

Set these to false in plugin_enabled when the corresponding opencode-tools plugin is loaded, to avoid duplicate sidebar panels.

MCP names truncate before the right-aligned status label. Every line remains within the 37-cell sidebar width and contains no trailing whitespace.

▼ MCP
-------------------------------------
• codegraph-global          Connected
• context7-global           Connected
• postgres-test-vendsystem   Disabled
• postgres-test-vendsystem…  Disabled
-------------------------------------

▶ MCP                           4/0/0
-------------------------------------

▶ MCP                           2/1/1
-------------------------------------

▶ MCP                           0/0/0
-------------------------------------

The collapsed summary shows success/warning/error counts: connected servers, disabled servers, and servers in any error state (failed, needs authentication, needs client registration, or unknown). Each number uses its bucket color — success, warning, or error — including when it is zero, and both separators are muted. For the empty 0/0/0 summary, each zero keeps its bucket color.

Context values come from the active session. The expanded panel uses Limit -, Tokens -, Used -, and Spent $0.00 when context values are unavailable; the collapsed summary uses -. The expanded Used value and collapsed summary are green below 40%, yellow from 40% through 60%, and red above 60%. Only a $0.00 value in the Spent row is muted.

▼ Context
-------------------------------------
Limit                            500K
Tokens                        322.12K
Used                              64%
Spent                           $0.00
-------------------------------------

▶ Context                         64%
-------------------------------------

LSP IDs stay in synchronized source order. Each row also shows the basename of the server's operating directory (root), right-aligned in the muted label color; a server with an empty root shows the id alone. Long IDs truncate with an ellipsis so expanded lines fit within 37 cells and collapsed lines fit within 36 cells.

▼ LSP
-------------------------------------
• typescript           opencode-tools
• yaml-ls              opencode-tools
-------------------------------------

▼ LSP
-------------------------------------
LSPs will activate as files are read
-------------------------------------

▶ LSP                               2
-------------------------------------

The collapsed count uses normal header text. Successful servers use a success bullet, failed servers use an error bullet, and unknown statuses remain present with a muted bullet. Header clicks affect only the current session selection.

TODO records stay in synchronized source order. Status markers occupy four cells, so wrapped content continues beneath the content column without trailing whitespace. TODO continuation lines align under the content column, and the collapsed summary rolls records into done/working/todo counts that exclude cancelled records.

▼ TODO
-------------------------------------
[✓] Explore existing panel patterns
[•] Implement synchronized TODO
    state and wrapped rows
[ ] Verify build and deployment
[-] Superseded task
-------------------------------------

▼ TODO
-------------------------------------
No TODOs for this session
-------------------------------------

▶ TODO                          4/3/2
-------------------------------------

These examples preserve the current SesTokens layout semantics from AGENTS.md. Fenced lines omit right-padding; OpenTUI flex alignment supplies the visual spacing at runtime. Values remain right-aligned within the 37-cell sidebar.

▼ SesTokens
------------------------------------
↻ turns                           97
↑ in                           4.41M
↓ out                         18.69K
▤ cache write                      0
▤ cache read                  24.77M
ø cache hit ratio              5.68×
✦ think                        2.87K
---                              ---
Σ total                       29.11M
------------------------------------

▼ SesTokens                    stale
------------------------------------
↻ turns                           97
↑ in                           4.41M
↓ out                         18.69K
▤ cache write                      0
▤ cache read                  24.77M
ø cache hit ratio              5.68×
✦ think                        2.87K
---                              ---
Σ total                       29.11M
------------------------------------

▶ SesTokens                   29.11M
------------------------------------

▶ SesTokens             stale 29.11M
------------------------------------

While the first snapshot loads, the expanded body and collapsed summary show Loading... without zero metrics. If the initial attempt and all three retries fail, they show Usage unavailable. A failed refresh after ready data retains the last successful snapshot as stale and recovers it to ready when a later refresh succeeds. The panel does not poll. Snapshots and collapse choices are memory-only, and the panel does not calculate cost.

These examples preserve every canonical SubAgent layout from AGENTS.md. Rows show the newest five direct children first; older entries appear in the separate Rest group. Fenced lines omit runtime right-padding and contain no trailing whitespace.

▼ SubAgent
------------------------------------
▶ SubAgent11 with super lo…   9m 45s
▶ SubAgent10                  1h 15m
▼ SubAgent9
  agent:                     general
  status:                    running
  time:                       15m 4s
  model:                 gpt-4o-mini
  Open Session
▶ SubAgent8                   2h 18m
▶ SubAgent7                   2h 18m
---                              ---
▼ Rest
▶ SubAgent6                   9m 45s
▶ SubAgent5                   1h 15m
▶ SubAgent4                      15s
▶ SubAgent3                      25s
▶ SubAgent2                       5s
▶ SubAgent1                    1h 2m
------------------------------------

▼ SubAgent
------------------------------------
▶ SubAgent11 with super lon
  g name that would normall
  y wrap but is too long to
   fit.
  agent:                     general
  status:                    running
  time:                       9m 45s
  model:                 gpt-4o-mini
  Open Session
▶ SubAgent10                  1h 15m
▼ SubAgent9                   15m 4s
▶ SubAgent8                   2h 18m
▶ SubAgent7                   2h 18m
---                              ---
▼ Rest
▶ SubAgent6                   9m 45s
▶ SubAgent5                   1h 15m
▶ SubAgent4                      15s
▶ SubAgent3                      25s
▶ SubAgent2                       5s
▶ SubAgent1                    1h 2m
------------------------------------

▼ SubAgent
------------------------------------
▶ SubAgent11 with super lo…   9m 45s
▶ SubAgent10                  1h 15m
▶ SubAgent9                   15m 4s
▶ SubAgent8                   2h 18m
▶ SubAgent7                   2h 18m
---                              ---
▼ Rest
▶ SubAgent6                   9m 45s
▶ SubAgent5                   1h 15m
▶ SubAgent4                      15s
▶ SubAgent3                      25s
▶ SubAgent2                       5s
▶ SubAgent1                    1h 2m
------------------------------------

▼ SubAgent                     stale
------------------------------------
▶ SubAgent11 with super lo…   9m 45s
▶ SubAgent10                  1h 15m
▶ SubAgent9                   15m 4s
▶ SubAgent8                   2h 18m
▶ SubAgent7                   2h 18m
---                              ---
▼ Rest
▶ SubAgent6                   9m 45s
▶ SubAgent5                   1h 15m
▶ SubAgent4                      15s
▶ SubAgent3                      25s
▶ SubAgent2                       5s
▶ SubAgent1                    1h 2m
------------------------------------

▼ SubAgent
------------------------------------
▶ SubAgent11 with super lo…   9m 45s
▶ SubAgent10                  1h 15m
▶ SubAgent9                   15m 4s
▶ SubAgent8                   2h 18m
▶ SubAgent7                   2h 18m
---                              ---
▶ Rest
------------------------------------

▶ SubAgent                     7/1/3
------------------------------------

▶ SubAgent               stale 7/1/3
------------------------------------

▼ SubAgent
------------------------------------
No subagents
------------------------------------

The collapsed summary reports successful/running/failed counts. A stale refresh retains the complete entry body and marks the panel stale; it does not replace rows with a loading message. Before a selected parent has ready data, the plugin emits no panel output.

Build the nine standalone minified ESM plugins and their imported shared artifact:

npm run build:plugins

Deploy to this repository's .opencode/ directory or the resolved global OpenCode config directory ($XDG_CONFIG_HOME/opencode, defaulting to ~/.config/opencode):

npm run deploy:local
npm run deploy:global

Each deploy command rebuilds first and automatically migrates managed configuration entries to the nine standalone entries in manifest order. It preserves unrelated plugin entries and preserves existing per-plugin options (quota and defaultState); quota options remain attached only to the quota entry. Local deployment also removes managed source entries from the project-root tui.json, because OpenCode loads it together with .opencode/tui.json; options in the selected .opencode config take precedence. Repeating either command produces the same files and configuration. Deployment does not edit plugin_enabled or disable the built-in MCP, LSP, or TODO panel. Set the overrides in the configuration example yourself when replacing any built-in panel. Fully restart OpenCode after deployment.

Deployment replaces stale managed SubAgent artifacts and removes stale managed source entries while preserving unrelated files and configuration entries.

The normalized quota runtime ID is now aamkye/opencode-tools-quota. This is an intentional ID change, so host-managed plugin state may reset during migration; the deployer still preserves quota's configuration options.

To remove the Context panel, remove ./opencode-tools-context.js from the plugin array and restart OpenCode. To remove the SesTokens panel, remove ./opencode-tools-ses-tokens.js from the plugin array and restart OpenCode. To remove the SubAgent panel, remove ./opencode-tools-subagent.js from the plugin array and restart OpenCode. To return to OpenCode's built-in MCP panel, remove ./opencode-tools-mcp.js from the plugin array, then remove the "internal:sidebar-mcp": false override (or set it to true) to re-enable internal:sidebar-mcp, then restart OpenCode. To return to OpenCode's built-in LSP panel, remove ./opencode-tools-lsp.js from the plugin array, then remove the "internal:sidebar-lsp": false override (or set it to true) to re-enable internal:sidebar-lsp, then restart OpenCode. To return to OpenCode's built-in TODO panel, remove ./opencode-tools-todo.js from the plugin array, then remove the "internal:sidebar-todo": false override (or set it to true) to re-enable internal:sidebar-todo, then restart OpenCode. To roll back the complete standalone migration, optionally restore the prior composed release and its configuration before restarting.

dist/
├── opencode-tools-shared.js
├── opencode-tools-home.js
├── opencode-tools-token-report.js
├── opencode-tools-context.js
├── opencode-tools-ses-tokens.js
├── opencode-tools-subagent.js
├── opencode-tools-quota.js
├── opencode-tools-mcp.js
├── opencode-tools-lsp.js
├── opencode-tools-todo.js
└── session-rename.ts

solid-js, @opentui/*, @opencode-ai/plugin, host SDK modules, and Node/Bun built-ins remain external and are provided by the OpenCode host.

The global session rename plugin is manual-only. It changes a session title only when the command is invoked. Use /session-rename Project planning notes to set an explicit, validated 3-8 word title. Use /session-rename without a title to generate one from recent user text and the latest selected user model. The plugin adds success or failure feedback as an ignored message and disables OpenCode's built-in title agent.

Build the plugin with npm run build:session-rename. Deploy it with npm run deploy:global; this installs dist/session-rename.ts as ~/.config/opencode/plugins/session-rename.ts. Deployment installs the new file before it removes the previously managed legacy artifact. Fully restart OpenCode after deployment.

Edit the relevant source, redeploy, then fully restart OpenCode to reload.

npm install       # install/refresh deps in node_modules
npm run typecheck # tsc --noEmit (informational; runtime resolves via Bun)
npm run build:plugins # rebuild all nine standalone plugins plus shared code
npm run deploy:local # rebuild and deploy into this repository
npm test          # run tests

This project was renamed to opencode-tools. Replace every prior project path, TUI entry, package name, token plugin filename, and build command with the paths shown above. Legacy files and aliases are intentionally not provided.

1. Reads the API key from the zai-coding-plan provider (falls back to ~/.local/share/opencode/auth.json, then ~/.config/opencode/auth.json and the older account.json locations).
2. Polls https://api.z.ai/api/monitor/usage/quota/limit every 10s (5min when the 5H quota is exhausted).
3. Renders bars + countdowns in the sidebar; expands for absolute counts.

1. Reads the OAuth access token from the openai provider entry in auth.json (also checks codex, chatgpt, opencode keys).
2. Extracts the chatgpt_account_id from the JWT for the ChatGPT-Account-Id header.
3. Polls https://chatgpt.com/backend-api/wham/usage every 10s (5min when the primary window is exhausted).
4. Renders the plan type and available primary/secondary quota windows with compact labels derived from each API-reported duration.

1. Sends the configured workspace credentials only to the fixed https://opencode.ai origin.
2. Reads quota data from the authenticated page's undocumented Solid hydration contract and fails closed when that contract changes.
3. Renders the rolling 5H, weekly 7D, and subscription month 1M windows with shared polling, countdown, reset, and a ten-minute stale horizon without exhausted backoff.

1. Reads assistant messages from opencode.db (SQLite, via bun:sqlite).
2. Aggregates token usage by model, provider, and session.
3. Resolves USD costs using a bundled models.dev pricing snapshot.
4. Formats a markdown report with summary, model breakdown, and top sessions.
5. Injects the report into the session via noReply prompt (no model invocation).