A context optimization and long-term memory skill that reduces LLM token usage via delta compression and intelligent caching.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install usewhisper
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install usewhisper using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Whisper Context?
Whisper Context is a high-efficiency utility designed to minimize API expenditure for AI agents by shrinking the context window sent to Large Language Models. By employing delta compression and intelligent caching strategies, this skill ensures that agents remain cost-effective while maintaining a persistent memory across multiple chat sessions. It acts as a sophisticated middle-layer that bridges the gap between ephemeral chat turns and long-term historical awareness for any agent in the Openclaw Skills ecosystem.
Built as a Node-based helper, it seamlessly integrates into existing developer workflows to provide tools for context retrieval, session ingestion, and advanced research. Beyond token savings, it provides robust Oracle search capabilities, allowing agents to perform deep research or complex planning without manual prompt engineering for history management.
Whisper Context Use Cases
- Reducing recurring API costs in long-running agent conversations by using delta compression.
- Maintaining user preferences and technical historical data across different chat sessions.
- Implementing advanced research and multi-step planning using the Oracle search functionality.
- Auditing token savings and monitoring performance through detailed cost and cache analytics.
How Whisper Context Works
- The agent executes the query_context command to retrieve a packed, compressed version of the conversation history based on the user and session IDs.
- The returned optimized context is injected into the model prompt, replacing the standard full chat history to save tokens.
- Once the LLM provides a response, the agent triggers the ingest_session command to persist the latest interaction into long-term memory.
- The system automatically manages delta compression and cache warming to ensure subsequent queries are both fast and inexpensive.
Whisper Context Setup
Install the skill directly into your workspace using ClawHub:
npx clawhub@latest install whisper-context
Configure the required environment variables where your agent can access them:
export WHISPER_CONTEXT_API_KEY=YOUR_KEY
export WHISPER_CONTEXT_PROJECT=openclaw-cost-optimization
# Optional: Defaults to https://context.usewhisper.dev
export WHISPER_CONTEXT_API_URL=https://context.usewhisper.dev
Whisper Context Data Schema & Taxonomy
The skill organizes data around specific projects, users, and sessions to ensure isolation and relevance. It tracks the following metadata taxonomy:
| Parameter | Description |
|---|---|
| user_id | A stable identifier for the end user to isolate personalized memories. |
| session_id | A unique identifier for the specific conversation thread or task session. |
| project | The organizational slug used to group related agent activities. |
| compression_strategy | The method used to shrink tokens, typically defaulting to delta. |
| memory_type | Classification for stored memories, such as preference or factual knowledge. |
Whisper Context Advanced Features
- Oracle Search: Perform deep-dive research and autonomous planning with configurable max steps.
- Delta Compression: Dramatically reduce input token counts by only processing the delta between turns.
- Cache Warming: Proactively load common queries into the L1 cache to minimize latency for frequent requests.
- Cost Summaries: Access detailed JSON reports on token usage and projected savings across specific timeframes.
- Stdin Support: Support for piping large JSON objects for session ingestion to avoid shell character escaping issues.
SKILL.md
Loading
Related Openclaw Skills
usewhisper-autohook
An automated memory management tool for OpenClaw agents that handles context retrieval and conversation ingestion via Whisper Context.
Gemini Voice Assistant
A voice-to-voice AI assistant leveraging the Gemini Live API for natural, real-time spoken interactions.
Gemini Assistant
A versatile multimodal AI assistant powered by Google Gemini, offering high-speed text and voice interaction capabilities.
Sentinel Security Monitor
Sentinel transforms any Android device running IP Webcam into a sovereign, frugal motion detection system integrated with AI agents.
API Design Reviewer
Automate RESTful API analysis, linting, and breaking change detection to ensure high-quality engineering standards.
auto-memory-pro
A curation layer for AI agent memory that transforms temporary patterns into permanent project rules and reusable skills.