The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →The fastest way to reduce OpenClaw token usage is to measure the context first, then apply the fix that matches the bloat: use /compact for oversized conversation history, cache-TTL pruning for old tool results, prompt caching for repeated stable prefixes, and smaller payloads or models when the request itself is too large. These controls solve different problems, so changing all of them blindly can remove useful detail without lowering the bill.
Antihost offers VPS plans with one-click OpenClaw, NVMe storage and a 14-day free trial.
What OpenClaw token usage includes
OpenClaw sends a model a request made up of more than your latest question. Depending on the session and configuration, the active context can include the system prompt, transcript history, tool definitions, tool results, skill descriptions, files, images and the new user message. The model also generates reply tokens. Input and output are counted according to the selected model and provider; token behavior is therefore model-specific.
As an Amazon Associate I earn from qualifying purchases.
A character count is only a rough indication. OpenClaw’s token-use reference notes that about four English characters per token is an average, not a billing formula. Code, structured data, non-English text, image inputs and long identifiers can tokenize very differently. Measure the actual request and response usage before estimating savings.
“Every model has a context window: the maximum number of tokens it can process.”
#1 Best Overall
Sale
TOZO A1 Wireless Earbuds Bluetooth 5.3 Light Weight in Ear Headphones | IPX5 Waterproof,2 Mics for AI Calls,Immersive Premium Sound Bass Headset with Charging Case,32 EQ via App Black
- [Ultra-Lightweight Ear Buds Designed for Small Ears] Each earbud weighs only 3.7g and features a compact, ergonomic in-ear design made especially for small ears. Secure, low-profile, and comfortable for workouts, all-day wear or overnight sleeping.The ultra-slim profile sits flush in your ear with no hard edges poking or rubbing, so you can even sleep on your side without discomfort.
- [Immersive Stereo Sound with TOZO OrigX Technology] TOZO OrigX tuning delivers clear vocals, balanced mids, and natural stereo sound for music, podcasts, and videos.
- [Long Battery Life for Daily Use] Get up to 7 hours of playtime on a single charge, with up to 32 hours total using the charging case—ideal for workdays, commuting, and extended listening sessions.
- [Bluetooth 5.3 & Stable Connection] Bluetooth 5.3 provides fast pairing, stable wireless performance, and reduced dropouts as you move around home or office.
- [Deep Bass with Clear Vocals] High-performance drivers produce punchy bass while keeping vocals clean and detailed for everyday listening.
OpenClaw project documentation, Compaction reference
Reducing token usage can mean reducing active-context size, avoiding repeated processing of an unchanged prefix, lowering generated output, or choosing a less expensive model for work that does not need the largest one. The right method depends on which component is dominant.
Measure the baseline before changing configuration
Start with one representative task or session. Record its token usage and inspect the context composition. Otherwise, a change such as pruning may appear ineffective because the real cost is a large system prompt, screenshots or verbose model replies.
Check reply usage with /usage tokens
Run:
/usage tokensThis exposes token usage on replies, allowing you to see whether a workload is producing unusually large input, output or both. Capture several normal turns rather than judging from a single unusual request.
Break down the active context with /context detail
Run:
/context detailThe detail view identifies contributors such as files, tool schemas, skills, the system prompt and transcript history. Look for the largest category:
- A large transcript points to compaction.
- Large, repeated tool results point to context pruning or output trimming.
- Large screenshots point to a lower image dimension.
- Extensive skill descriptions or tool schemas point to shorter definitions or fewer enabled capabilities.
- High reply usage points to shorter task instructions, tighter output limits where available, or a smaller model for exploratory work.
Use /context map for a visual view
Run:
/context mapThe map presents the context as a treemap, which can make a single oversized file, tool response or image contribution obvious. Use it when the text breakdown is difficult to scan or when several sources appear similar in the detail view.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteRank #2
- 【IP68 Waterproof Multifunctinal Smart Watch】Torjalph fitness activity tracker smart watch is IP68 waterproof technology ,you can wear it when swimming,handwashing,it's rain-proof which means that you perform multiple exercise modes outside without worrying about sweat or rain corroding your fitness smart watch,and this watch equipped with a 230 mAh large capacity battery magnetic charging ,and this sports activity tracker watch can be used 5-7 days If fully charged.
- 【Larger 1.83'' Full Screen Watch & Multi-dial Selection】1.83 inch full HD color touch screen smart watch for men / women provides excellent 240*284 resolution image quality ,This Bluetooth smart watch offers a better touch experience and remarkable immersion and ,more than 200 watch faces for you to choose from App"Da Fit ", you can change the dial style according to your mood to wear keeping the world on your wrist
- 【Accurate Activity Tracker Smart Watch 】sports watch activity tracker uses sweat and dust resistant breathable silicone strap, our Fitness Tracker Smart Watch for Android iOS Phones supports 12 sports modes,walking,running,bicycle,swimming,badminton,basketball,football,rope skipping. This bluetooth calls smart watch for men women is lightweight and stylish, bringing more convenient and relaxing sports experience while also providing more protection.
- 【Your 24 Hour Personal Daily Health Monitor】Fitness watch for women men full touch screen IP67 waterproof bluetooth smartwatch for Android iOS Phones with heart rate blood pressure monitor multifunction Andriod smart watch with steps counter sleep monitor sports fitness watch, monitoring blood oxygen saturation allows you to understand your physical condition in real time to track fitness progress ,to know your health status in time.
- 【More Functions Bluetooth Call Smart Watch】This fitness activity smart watch full screen for iPhone and Android phones includes a bunch of new features including:The bluetooth smart watch can hang up the phone, receive and display WhatsAPP /Text /Facebook /Twitter /Messenger reminder.Support sedentary reminder, alarm clock, stopwatch, timer,weather information,music control, remote camera control compatible with Android 5.0 and above or iOS 9.0 and above smartphones.
Use /compact when conversation history is the problem
Compaction summarizes older conversation turns into a persisted compaction entry while retaining the recent tail. It is the appropriate intervention when the transcript itself has grown too large over many turns.
How compaction changes a session
OpenClaw chooses a compaction boundary and preserves matching assistant tool calls with their tool results when selecting that boundary. The result is a shorter active history represented by a summary plus recent messages. Because the compaction entry is persisted, later turns continue from the summarized state rather than rebuilding the entire old transcript each time.
Compaction is not the same as deleting a session. The recent tail remains available, and the summary carries forward the important older state. However, details that were not included in the summary may no longer be immediately recoverable in the active context. If an exact command, identifier or quotation matters, place it in the recent tail or restate it after compacting.
When to compact
- Use it after a long planning or debugging session in which old dialogue is the largest item in
/context detail. - Use it before starting a new phase of work that still needs the session’s decisions but not every exploratory turn.
- Do not use it as the first response to a tool-output problem; pruning or trimming is more targeted when tool results dominate.
A safe compaction procedure
- Run
/context detailand confirm that transcript history is a major contributor. - Save any exact values you must preserve, such as environment names, file paths, API parameters or acceptance criteria.
- Run
/compact. - Ask a short verification question or inspect the next context view to confirm that the summary retains the project goal and current decisions.
- Continue the session, supplying any exact detail that the summary omitted.
Use cache-TTL pruning for stale tool results
Tool calls can return large search results, file listings, logs or execution output. Keeping every old result in the active request increases input tokens even when the conversation itself is still useful. Session pruning removes old tool-result context from the projected in-memory request while leaving the full session transcript available for inspection.
How cache-TTL pruning works
With contextPruning.mode set to cache-ttl, OpenClaw waits for the configured cache time-to-live and context-size conditions before pruning eligible old tool results. This delay prevents immediately discarding material that may still be reused during an active burst of work. The transcript remains on disk, so pruning does not rewrite the conversation the way compaction does.
Example configuration
agents:
defaults:
contextPruning:
mode: "cache-ttl"
ttl: "1h"The exact pruning behavior depends on the OpenClaw version and the provider route. Confirm the supported keys and current semantics in the documentation before deploying a shared configuration.
Rank #3
- [Low Noise High Fidelity] New bee guitar cord with 22AWG oxygen-free-copper center conductors provides enhanced signal transfer, the PVC insulation material and the high-density stranded copper double shield provide noise rejection for quiet operation
- [Ideal for Choice] New bee 1/4 straight and right angle metal connectors, ensures a secure connection in any instrument, pedal, or amp jack. L shape great fro on stage guitar stands
- [Wide Compatible] Suitable for mixer, power amplifier, microphone, effector, decoder, equalizer, electoronic organ, electric guitar, electric bass and other electronic equipments. Widely used in the installation of smart home, karaoke speaker, professional stage audio, etc
- [Easy Disassemble] Making incidental repairs quick and easy! Very nice, flexible cable
- [Dont Hesite to Order] 24 hours customer services and professional technology team are standing by
Pruning versus compaction
| Technique | Main reduction target | What remains | Main trade-off |
|---|---|---|---|
/compact |
Older conversation text | A persisted summary and recent tail | Some detail is represented only in the summary |
cache-ttl pruning |
Old tool-result context | The full transcript on disk | Pruning waits for TTL and context-size conditions |
| Prompt caching | Repeated processing and cache writes for a stable prefix | The cached prefix and usage counters | Provider route and request-history stability are required |
| Tool/output trimming | Oversized tool results | Only the retained output | Less raw detail is immediately available |
| Smaller model selection | Cost of exploratory turns | The selected model’s output | Quality and capability may differ |
Use compaction when history is the problem and pruning when old tool output is the problem. They can be combined: prune transient results while retaining a compacted summary of the conversation.
Improve prompt-cache reuse for stable workloads
Prompt caching reuses an unchanged prompt prefix instead of processing that prefix from scratch on every request. It can reduce repeated input processing and cache-write work, but it does not make a changing transcript free. Cache effectiveness depends on the provider route, model and how much of the request stays identical.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Configure cache retention
OpenClaw exposes cacheRead and cacheWrite usage and supports retention values such as none, short and long. Retention can be set globally and overridden per model or per agent. A longer retention period can help an always-on workflow with a stable prefix, while it may add retention cost or provide little value for irregular sessions.
Keep a cache warm with a heartbeat
If a provider’s cache window lasts one hour, a heartbeat interval just below that window can keep the prefix warm during idle periods. OpenClaw’s example uses a 55-minute heartbeat for a one-hour cache window:
agents:
defaults:
params:
cacheRetention: "long"
contextPruning:
mode: "cache-ttl"
ttl: "1h"
heartbeat:
every: "55m"This is an example, not a universal interval. Check the provider’s cache window and whether heartbeat requests use the same route, model and prefix. If any of those change, the expected cache hit may disappear.
When caching is not the right fix
- A transcript that changes substantially every turn may have little reusable prefix.
- Provider routes that do not support the relevant cache controls cannot use this optimization.
- Caching does not remove oversized tool results or shorten a generated answer.
- A very short or infrequent session may not justify a long retention setting or heartbeat traffic.
Reduce the payload sent to the model
After identifying the largest contributors, reduce the input itself. These changes are generally more predictable than trying to tune percentages because they remove tokens before the request is sent.
Rank #4
- Digital Family Calendar: BSIMB digital calendar wall touch screen puts an end to your chaotic life. Designed for busy families, the digital calendar keeps you organised and in step with your family. Assign different colours and tasks to each family member to make family scheduling easier and more intuitive.
- Easy Setup and Auto-Sync: Plug in, connect to WLAN and auto-syn your electric calendar in minutes. Digital wall calendar compatible with Google calendar, iCloud calendar, Outlook calendar, Cozi calendar and Yahoo calendar. Add events directly on the device or through the ecalendar app. Electronic calendar stays current whether you're at work or traveling—no manual updates needed.
- All-In-One Home Assistant: Smart calendar builds habits and encourage independence with personalised chores. Display daily meal plan to reduce the daily ‘What's for dinner?’ stress. To-do lists keep track of everything you want. Turning chores into valuable achievements with stars and rewards, making life easier and more enjoyable.
- Manage Anytime Anywhere: BSIMB electronic calendar for family stay connected with the powerful ecalendar app. Access and manage your digital wall calendar anytime, anywhere with free features like stars & rewards, photo screensaver, chore charts, meals plans, to-do lists etc. Electronic planner paid features like Magic Import tailor advanced tools for users with special needs.
- Stay in Touch with Loved Ones: Share photos or videos from your phone to electronic wall calendar with app, instantly update your life moment, and keep in touch with your family, especially those who are far away; you can also surprise your smart calendar digital wall by sending him a wish card; it's also great for keeping tech-illiterate elders in touch with the youngsters.
Trim tool outputs at the source
Configure search, file and execution tools to return only the fields and rows needed for the next decision. Prefer bounded results, summaries, pagination and targeted file ranges over entire logs or directories. If a command can emit machine-readable data, select only the relevant properties rather than returning a full record. Keep a path or query that can retrieve the omitted detail again when necessary.
Do not silently truncate output that the model must verify. A shorter result is useful only when the retained portion still contains the status, errors, identifiers or lines required for the task. If a tool’s output is routinely enormous, redesign the tool call or wrapper instead of relying solely on later pruning.
Lower screenshot dimensions
Screenshot-heavy sessions can add substantial image input. OpenClaw’s token-use guidance recommends lowering agents.defaults.imageMaxDimensionPx when full-size screenshots are unnecessary. Use a dimension that preserves the UI text or visual feature the model must inspect; reducing it too far can force additional screenshots and erase the saving.
Keep skill descriptions concise
Skill descriptions and instructions are part of the context when those skills are available to the agent. Remove duplicated explanations, obsolete examples and long prose that does not affect task selection. Put detailed material behind an on-demand reference rather than loading it into every request, while retaining the constraints the model must always follow.
Recommended Free Tools
Select a smaller model for exploratory work
OpenClaw recommends preferring smaller models for verbose exploratory tasks when quality permits. Use a capable model for requirements that need stronger reasoning, coding accuracy or specialized knowledge, and a smaller one for routine classification, drafting, formatting or first-pass exploration. Compare the resulting error rate and rework, not just the price of one request: a cheap answer that requires several retries may cost more overall.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A practical reduction workflow
- Choose a representative workload. Include the long-running session, tool calls, images and model you actually use; a tiny test will not reveal production bloat.
- Measure replies. Run
/usage tokensover several turns and note input and output patterns. - Map the context. Use
/context detail, then/context mapif the largest contributor is unclear. - Classify the dominant source. Decide whether it is transcript history, stale tool results, repeated stable prefix, images, skills, tool schemas or generated prose.
- Apply one targeted change. Compact history, enable cache-TTL pruning, trim outputs, lower image dimensions, shorten skill descriptions or select a smaller model.
- Repeat the same measurement. Compare equivalent turns, including answer quality and any extra retries.
- Document the trade-off. Record what detail is no longer immediately available, how long pruning waits, which provider route supports caching and which tasks remain on the larger model.
Common failure modes and fixes
“I compacted, but usage is still high.”
Compaction only targets older conversation text. Re-run /context detail. If tool results, screenshots or skill descriptions now dominate, use pruning or payload trimming instead.
“Pruning removed information I still need.”
The full transcript remains available, but the old tool result is no longer projected into the active request. Re-run the tool with a narrower query, retrieve the needed file range again, or preserve a concise summary in the conversation before the result becomes eligible for pruning.
“The cache setting produces no visible saving.”
Check that the provider route and model support caching, that the prefix is genuinely unchanged, and that requests arrive within the provider’s retention window. Inspect cacheRead and cacheWrite counters rather than assuming a cache hit.
“A smaller model costs more in practice.”
Measure the complete workflow, including retries, corrections and human review. Keep the larger model for tasks where a quality failure is expensive, and use the smaller model only where its output meets the acceptance criteria.
“Lowering screenshots made the answers worse.”
Increase agents.defaults.imageMaxDimensionPx enough to preserve the text or controls under inspection, or capture a cropped region instead of repeatedly sending a low-resolution full screen.
How to choose the least disruptive intervention
| If your measurement shows | Start with | Why |
|---|---|---|
| Transcript history is largest | /compact |
It replaces old dialogue with a persisted summary and keeps the recent tail. |
| Old search, file or execution results are largest | Cache-TTL pruning | It removes stale tool-result context without deleting the full transcript. |
| The same system and task prefix repeats across idle gaps | Prompt caching and a suitable heartbeat | It can reuse the stable prefix when the provider route supports it. |
| Screenshots, files or tool responses are oversized | Payload trimming | It reduces tokens before they enter the request. |
| Exploratory replies are verbose and quality requirements are modest | A smaller model | It targets model processing cost while preserving the task workflow. |
There is no authoritative universal percentage reduction for these settings. Savings vary with session length, provider pricing, cache behavior, model tokenizer, image content and how often the workload repeats. Treat every percentage claim as workload-specific and validate it with your own equivalent requests.
Quick Recap
Operational checklist
- Run
/usage tokensbefore and after changes. - Use
/context detailto identify the actual source of growth. - Use
/context mapwhen a treemap will reveal an outlier faster. - Compact when old conversation text is the main problem.
- Prune with
cache-ttlwhen stale tool results are the main problem. - Keep tool output bounded and recoverable.
- Lower screenshot dimensions only as far as the task allows.
- Keep skill descriptions short and non-duplicative.
- Use cache retention and heartbeats only for stable prefixes and supported provider routes.
- Assign smaller models to tasks whose quality criteria they can meet.
- Recheck configuration keys and provider support against the current OpenClaw documentation before rollout.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches
