Skip to content
Token Optimizer

OpenClaw

OpenClaw runs Token Optimizer as a TypeScript plugin with the deepest waste-analysis surface of any platform. It carries features that exist nowhere else, including a 16-detector audit, an 8-tab dashboard, and progressive checkpoints across fill and quality bands.

Supported surfaces

Section titled “Supported surfaces”

OpenClaw has one surface, the plugin. Install it from GitHub, ClawHub, or source. See the OpenClaw install page. The cross-platform grid is at /reference/capability-matrix/.

Features unique to OpenClaw

Section titled “Features unique to OpenClaw”

These exist on OpenClaw and not on Claude Code.

  • 16 waste detectors, against 11 on Claude Code and Codex
  • An 8-tab dashboard: Overview, Context, Quality, Waste, Agents, Sessions, Daily, Manage
  • Progressive checkpoints at fill bands 20/35/50/65/80 percent, quality drops at 80/70/50/40, and milestones
  • Read-cache with warn and block modes plus a .contextignore file
  • Git-aware context suggestions from the working-tree state
  • Per-turn token breakdown with cache analysis across all providers
  • Agent cost breakdown that separates orchestrator from worker
  • Model switch simulation that estimates savings from routing to a cheaper model
  • Drift detection, impact validation, and distortion-bounds analysis
  • Local checkpoint telemetry, opt-in through TOKEN_OPTIMIZER_CHECKPOINT_TELEMETRY=1

Model-agnostic gateway

Section titled “Model-agnostic gateway”

OpenClaw supports more than 30 models across Claude, GPT-5, Gemini, DeepSeek, Qwen, Mistral, Grok, and local models via Ollama. Cost tracking carries four pricing tiers: Anthropic direct, Google Vertex AI global, Vertex AI regional with a 10 percent uplift, and AWS Bedrock. Pricing overrides go in openclaw.json. See the configuration reference.

Not ported from Claude Code

Section titled “Not ported from Claude Code”

These features are not in the OpenClaw plugin. Each is blocked on a plugin-API surface OpenClaw does not yet provide.

  • Quality Nudges and Loop Detection, which need a session-visible notification surface
  • Bash output compression, which needs a tool-input mutation hook
  • Context Intel Promotion, the post-compaction tool-output digest
  • MCP tool introspection, which is Claude Code specific
  • Live quality bar status line
  • Memory Health audit, Attention Optimizer, the JSONL trim and dedup tools, routing injection, and tool-result archive
  • Keep-Warm automation

Runtime

Section titled “Runtime”

OpenClaw exposes its own CLI through npx token-optimizer. There is no TOKEN_OPTIMIZER_RUNTIME prefix.

Terminal window
npx token-optimizer audit --days 30
npx token-optimizer quality
npx token-optimizer dashboard

Inside OpenClaw, run /token-optimizer for a guided audit with coaching.

Doctor command

Section titled “Doctor command”
Terminal window
npx token-optimizer doctor --json

Dashboard

Section titled “Dashboard”

OpenClaw ships its own 8-tab TypeScript dashboard, separate from the shared Python dashboard. It auto-regenerates on session end. Open it manually with npx token-optimizer dashboard. There is no daemon port in the shared port map, because the dashboard is its own build.