ContextZip is a lightweight prompt framework that compresses prompts, conversations, technical logs, and documents while preserving reasoning-critical information.
Instead of generating traditional summaries, ContextZip focuses on keeping the information that affects future reasoning and decision-making while removing unnecessary token consumption.
Large Language Models often waste context window capacity on:
- Repeated information
- Conversational filler
- Emotional language
- Outdated context
- Verbose descriptions
As conversations become longer, token costs increase and model performance may degrade due to context dilution.
ContextZip solves this problem by extracting only the information required for future reasoning.
Transform verbose prompts into compact and structured instructions.
Convert long chat histories into reusable memory blocks.
Extract errors, root causes, affected components, and troubleshooting steps.
Turn large project descriptions into concise technical context.
Reduce lengthy documents into actionable summaries while preserving key facts.
I am currently maintaining a Weaver E9 deployment running on TongWeb with DM Database. Recently the portal news module started throwing exceptions. I checked the server resources and database connectivity and both appear healthy. I need help identifying the root cause.
Task: Investigate Weaver E9 portal news exception
Environment:
- Weaver E9
- TongWeb
- DM Database
Verified:
- Database connection OK
- System resources normal
Issue:
- DMException: String conversion error
Estimated Reduction: 80%
Always keep:
- Objectives
- Constraints
- Technical facts
- Dates
- Numbers
- Decisions
- Dependencies
- Open issues
Always remove:
- Greetings
- Emotional language
- Repetition
- Filler content
- Irrelevant history
Represent information using:
- Bullet lists
- Key-value pairs
- Hierarchical summaries
Use the smallest possible token footprint without changing meaning.
You are ContextZip, a loss-aware context compression engine.
Goal:
Compress user input into the smallest possible token footprint while preserving all information necessary for future reasoning.
Rules:
- Remove greetings, filler words, and emotional language.
- Remove duplicate information.
- Merge similar statements.
- Preserve:
- Objectives
- Constraints
- Key facts
- Dates
- Numbers
- Technical details
- Decisions made
- Outstanding problems
- Use bullet points whenever possible.
- Do not answer the user's question.
- Only output the compressed context.
Output Format:
Task:
Context:
- Key information
Constraints:
- Limitations or requirements
Known Facts:
- Verified information
Open Issues:
- Current problems
Compression Estimate:
- Estimated reduction percentage
| Content Type | Typical Reduction |
|---|---|
| Prompts | 50% - 80% |
| Conversations | 80% - 95% |
| Technical Logs | 70% - 90% |
| Documents | 60% - 90% |
Actual results depend on content quality and redundancy.
- AI Agents
- Long-Term Memory Systems
- RAG Pipelines
- Customer Support Bots
- Coding Assistants
- DevOps Assistants
- Enterprise Knowledge Bases
- Research Workflows
- Multi-level compression modes
- Context quality scoring
- Prompt benchmarking
- Agent memory optimization
- Automatic token reduction reports
The goal is not to summarize.
The goal is to preserve reasoning while minimizing tokens.
MIT License