whisper-context

What This Skill Does

The whisper-context skill is a powerful performance optimization layer for OpenClaw agents, designed specifically to reduce API expenditure by implementing advanced context management. In the era of large-scale LLM operations, the cost of repeatedly re-sending chat history in the context window is a significant financial burden. Whisper Context solves this by utilizing delta compression and intelligent caching. Instead of sending raw history, the skill retrieves a compressed, optimized version of the session state. It acts as an abstraction layer between your agent and the LLM, enabling persistence that transcends individual chat sessions through integrated long-term memory functions. Whether you are managing complex multi-turn conversations or building agents that require deep recall over long durations, this skill ensures that your context window remains lean, fast, and cost-effective.

Installation

To begin, ensure you have the OpenClaw environment initialized. Installation is handled through the ClawHub ecosystem. Run the following command in your terminal to fetch the skill and install it into your local workspace:

npx clawhub@latest install whisper-context

This command automatically places the necessary Node.js helper files into your ~/.openclaw/workspace/skills/ directory. Once installed, you must configure your environment variables to point the agent to the context service. Set WHISPER_CONTEXT_API_KEY to your credentials, and ensure WHISPER_CONTEXT_PROJECT is set to your preferred project name. If the project identifier you provide does not exist, the system will initialize it automatically during the first authenticated request.

Use Cases

• Long-term Agent Memory: Ideal for personal assistants that need to recall user preferences or historical decisions across weeks or months of interaction.
• High-Volume Context Compression: Essential for agents operating on expensive models where every token saved significantly impacts the monthly burn rate.
• Stateful Conversational Flows: Perfect for complex workflows where the current state of a task must be preserved across multiple independent process invocations.
• Cost-Optimized Analytics: Provides built-in hooks to monitor your usage patterns and cost metrics directly through the command line.

Example Prompts

1. "What is the current status of the project requirements discussed in last Tuesday's session?"
2. "Summarize all previous decisions regarding the UI architectural changes we made in the last three sessions."
3. "Save this session summary: the user confirmed the migration to the new cloud provider and requested a follow-up meeting on Friday."

Tips & Limitations

To achieve maximum token savings, always prioritize calling the query_context method before initiating your main LLM prompt. By default, the delta compression strategy is optimized for performance; avoid overriding this unless you have a specific requirement for full-text transmission. Ensure that your user_id and session_id are derived from your agent's internal logic rather than static strings, as these are the primary keys for cache invalidation and memory retrieval. Be aware that reliance on network-based context services introduces a small latency overhead; however, this is almost always offset by the reduction in model processing time.