OpenClaw Token Optimization: Bootstrap Budgets and Gateway Cost
A source-backed engineering report on OpenClaw’s persistent-gateway token economics: continuation-skip injection, bootstrap caps, startup context, image budgets, and per-agent profiles.
The highest-value controls are visible in the docs.
The report keeps only source-backed metrics and treats savings as workload-dependent unless the cited source gives a number.
Skips workspace bootstrap reinjection on safe continuation turns.
Default per-file workspace bootstrap cap in characters.
Default total bootstrap cap across workspace files.
Default longest-side image cap before provider calls.
Token waste usually comes from repeated context, not one bad prompt.
OpenClaw is a persistent gateway, so token spend can come from chat turns, channel routing, heartbeat or startup runs, memory work, tool calls, image processing, skills, and delegation.
Short durable context, scoped tools, measured output, and deliberate compaction beat blanket context hoarding.
Always-on instructions
Large memory files and broad project docs are paid for repeatedly.
Scoped surfaces
Expose only the tools, files, and context needed for the current workflow.
Measure first
Usage commands and context inspection prevent guesswork.
Context Injection Is the Primary Switch
The default is always. continuation-skip changes repeated context behavior.
OpenClaw documents agents.defaults.contextInjection as always by default. continuation-skip skips workspace bootstrap reinjection on safe continuation turns while preserving rebuilds for heartbeat runs and post-compaction retries.
The practical levers are small enough to operate.
Skips bootstrap reinjection on safe continuation turns.
Cuts repeated static context in normal back-and-forth sessions.
Disables workspace bootstrap and context-file injection every turn.
Use only for specialized agents with custom context engines.
Default 20,000 chars per bootstrap file.
Prevents one large workspace file from dominating the prompt.
Default total bootstrap cap across workspace files.
Do not convert this into a universal savings percentage. The report keeps the number tied to its original source and context.
Controls map to operating decisions.
| Lever | Documented behavior | Operating decision | Source |
|---|---|---|---|
| contextInjection: continuation-skip | Skips bootstrap reinjection on safe continuation turns. | Cuts repeated static context in normal back-and-forth sessions. | [openclaw-config] |
| contextInjection: never | Disables workspace bootstrap and context-file injection every turn. | Use only for specialized agents with custom context engines. | [openclaw-config] |
| bootstrapMaxChars | Default 20,000 chars per bootstrap file. | Prevents one large workspace file from dominating the prompt. | [openclaw-config] |
| bootstrapTotalMaxChars | Default 60,000 chars across bootstrap files. | Caps total workspace prelude cost. | [openclaw-config] |
| startupContext limits | Startup context has separate budgets. | Keep daily memory startup excerpts bounded. | [openclaw-config] |
Lowest-waste default config
The operating pattern keeps useful context close and noisy context out of the recurring prompt path.
Inspect context first
Run /status, /usage, /context list, and /context detail before changing caps.
Shorten bootstrap files
Keep bootstrap files short and move detailed logs into retrievable memory files.
Use agent profiles
Use per-agent profiles: lean workers, richer main agent, minimal background jobs.
Disable unused skills
Disable unused skills and keep skill descriptions concise.
Lower image budgets
Lower image dimensions and quality for image-heavy runs.
Claims Removed or Downgraded
The older report relied on exact baseline prompt-token estimates, old issue numbers, and broad savings percentages that were not preserved as current facts.
No fabricated stats, no fake case studies, no universal savings claims without a source-backed measurement path.
[openclaw-config]
OpenClaw agent configuration
[openclaw-context]
OpenClaw context concepts
[openclaw-token-use]
OpenClaw token usage reference
Make context injection a visible cost decision.
Use the documented controls for bootstrap caps, continuation behavior, memory, skills, image budgets, and agent profiles before scale turns context into recurring spend.
A practical report for teams operating OpenClaw as a persistent gateway.
Make the hidden context surface operable.
The full report gives operators a clear path from configuration to measured behavior.
- Context injection and bootstrap controls
- Memory, skills, and image-budget guidance
- Decision matrix for agent profiles