Skip to content

Capability matrix

Token Optimizer runs on nine surfaces. Feature coverage varies because each host exposes a different hook surface. The grid below shows exactly what works where. The full feature set lands on Claude Code; other platforms cover what their host APIs allow.

Legend: Yes = full support, Part = partial or capability-gated, Beta = beta status, No = not available, n/a = not applicable, verify = not confirmed in source.

Each runtime serves its dashboard on a fixed port, so multiple runtimes run side by side without conflict.

PlatformPortURL
Claude Code (CLI and VS Code)24842http://localhost:24842/token-optimizer
Codex (CLI and Desktop)24843http://localhost:24843/token-optimizer
Hermes24844http://localhost:24844
Copilot24845http://localhost:24845/token-optimizer (verify)
CapabilityClaude Code CLIClaude Code VS CodeCodex CLICodex DesktopOpenCodeOpenClawHermesCopilot CLI (beta)Copilot VS Code (beta)
Quality scoring7 signals7 signals7 signals (GPT-5.x curves)7 signals7 signals (MRCR curves)7 signals (two-stage)3 signals3 signals3 signals
S/A/B/C/D/F gradesYesYesYesYesYesYesYesYesYes
Smart Compaction (PreCompact/PostCompact)YesYesApprox. via prompt + StopApprox. via prompt + StopYes (mode-aware)Yes (progressive)No (nudge only)No (server-side)No (server-side)
Session continuity / checkpointsYesYesYesYesYesYes (progressive)Yes (via history db)Part (session-start restore)Part (session-start restore)
Status line (terminal quality bar)YesVia VSIXNoNoNoNoNoNoNo
Dashboard (visual HTML)Yes (24842)Yes (24842)Yes (24843)Yes (24843)YesYes (8-tab)Yes (24844)Yes (24845, verify)Yes
Daemon (bookmarkable URL)YesYesYesYesverifyNo (own build)Via CLI subcommandverifyverify
Keep-Warm automationYes (macOS, opt-in)YesNoNoNoNoNoNoNo
Bash output compressionYesYesExperimental opt-inExperimental opt-inNoNoNoPart (gated on CLI ≥1.0.24)No
Delta read substitutionYesYesNoNoNoNoNoNo (blocked #2585)No (blocked #2585)
Structure-map substitutionYesYesNoNoNoNoNoNo (blocked #2585)No (blocked #2585)
Per-prompt quality steeringYesYesVia UserPromptSubmitVia UserPromptSubmitYesverifyNoNo (regressed #3727)No (regressed #3727)
Cache-health watchdogYesYesPartPartPartPartPartPartPart
Cost trackingYes (Claude)YesYes (GPT-5.x)YesYes (30+ models)Yes (30+ models, 4 tiers)Yes (model-agnostic)Yes (AI Credits)Yes (per-request nanoAIU)
Fleet AuditorYesYesYes (cross-system)YesNo (verify)Via own dashboardNoNoNo
Token CoachYesYesYes (AGENTS.md)YesverifyYesNoNoNo
Memory / config auditYesYesYes (AGENTS.md + memories)YesNo (verify)YesNoNoNo
Read-cache (warn/block modes)No (verify)NoNoNoNoYesNoNoNo
Git-aware contextNo (verify)NoNoNoNoYesNoNoNo
Drift detectionNo (verify)NoNoNoNoYesNoNoNo
Impact validationNo (verify)NoNoNoNoYesNoNoNo
Agent cost breakdownNo (verify)NoNoNoNoYesNoNoNo
Model-switch simulationNo (verify)NoNoNoNoYesNoNoNo
Proactive context nudgeYesYesYesYesYesNo (verify)Yes (at 70%)Part (version-gated)No
Crash recoveryYes (StopFailure)YesApprox. (Stop + compact)Approx.verifyverifyverifyYes (in-flight tally)Yes (debug-log source)

The repeated No entries fall into a few clear groups.

Read interception (delta and structure map) needs a hook that can rewrite the result of a file read before the model sees it. Claude Code supports this; Codex lacks updatedInput on its Read hook; Copilot’s additionalContext on preToolUse is broken upstream (issue #2585). OpenCode and OpenClaw have no equivalent hook in their plugin APIs.

Bash output compression needs a hook that can rewrite tool input silently. Available on Claude Code, experimental opt-in on Codex, gated on a Copilot CLI version, and unavailable on the TypeScript platforms and Hermes.

Quality Nudges and Loop Detection need a session-visible notification surface to inject a warning the model can act on. Claude Code has it via UserPromptSubmit; OpenCode, OpenClaw, and Hermes do not yet expose one.

Status line needs a terminal status-bar surface. Claude Code has it (and the VS Code VSIX surfaces it in the editor). No other host exposes one.

Keep-Warm is Claude Code only, and only on API billing. OpenAI caching is automatic with no cache-write premium to recover; the other hosts have no comparable lever.

OpenClaw extras (read-cache warn/block, git-aware context, drift, impact validation, agent cost breakdown, model-switch simulation) are features of the native OpenClaw TypeScript engine. Several of these have no equivalent in the shared Python engine yet, marked (verify) for Claude Code until confirmed.

Where a platform exposes 7 signals, the score uses the full set: context-fill degradation, stale reads, bloated results, duplicate system reminders, compaction depth, decision density, and agent-dispatch efficiency. Hermes and Copilot fall back to a 3-signal subset because their session schemas do not persist the other signals reliably. The S/A/B/C/D/F grade scale is identical everywhere, so grades are comparable across platforms even when the underlying signal count differs.

For install commands, doctor commands, hook profiles, and platform-specific caveats, see the platform pages: Claude Code, Codex, OpenCode, OpenClaw, Hermes, and Copilot.