What percentage of OpenClaw tokens are wasted on context?

Analysis shows 93.5% of tokens in multi-message conversations are spent on static workspace content that never changes, representing massive waste on repeated context injection.

Why does OpenClaw use 40K tokens per message?

OpenClaw injects workspace files into the system prompt on every single API call. Even a simple 'yo' message consumes ~38K-43K tokens because the full workspace context is re-sent every time.

How much does OpenClaw token usage cost per month?

Costs vary widely: basic setups cost $6-13/month, small business workflows run $25-50/month, and heavy automation with premium models can exceed $200/month. One user reported $3,600/month with 1.8M tokens.

How to reduce OpenClaw token costs?

Use cheaper models (GPT-5 Nano or Gemini 3 Flash at ~$1/1000 messages), reduce workspace file injection, lower heartbeat frequency, and consider third-party optimizers like Lossless-Claw that claim 70% token reduction.

What is the 150K token system prompt issue in OpenClaw?

One fork investigation found OpenClaw sending ~166K input tokens per API call to Gemini 3.1 Pro on the first turn, with only 29 output tokens. The prompt includes hardcoded sections, skill prompts, bootstrap files, and tool JSON schemas.

Will OpenClaw fix the token waste problem?

GitHub Issue #59427 proposes lazy-loading context only when needed and a 'lightweight mode' config option. No official fix has been released yet, though the community has proposed several optimization approaches.

← Back to dashboard

clawsmith.com/signal/openclaw-baseline-40k-tokens-per-message-workspace-injection

⚠ IssueUnderservedCoreLive

OpenClaw Baseline Token Bloat: 40K Tokens Per Message From Workspace File Injection

Every OpenClaw message incurs ~40K tokens of context loading regardless of reply length. Workspace files are re-injected into the system prompt on every API call, causing 93.5% token waste in multi-message conversations. Users report bills from $3,600/month (1.8M tokens). One system was observed sending 166K input tokens per call. No lazy-loading or first-message-only option exists.

Product Idea from this Signal

A plugin that prunes OpenClaw's 40K-token workspace injection down to only the files relevant to each message, cutting baseline API costs by 60-80%

217 ▲

OpenClaw injects the entire workspace context (SOUL.md, agent configs, skill descriptions, tool declarations) into every single message, consuming 40,000+ tokens before the user even types a word. This fixed overhead makes smaller models unusable (they hit context limits immediately) and multiplies API costs 3-5x for conversational sessions. This plugin intercepts the context assembly step and applies relevance filtering, only including workspace files and tool declarations that match the current message's intent, dropping the baseline from 40K to under 10K tokens.

PLUGINOPEN-SOURCECOST-OPTIMIZATIONCONTEXT-MANAGEMENT

CompetitiveView Opportunity →