Cache-Assembler: Optimizing Anthropic API Costs via Prompt Caching

Research context and background

Cache-Assembler is a lightweight software tool designed to help developers save money when using the Anthropic Messages API. It works by making sure that parts of user prompts stay exactly the same across different requests. This allows Anthropic's built-in prompt-cache system to work properly. Without this tool, many small changes in how data is sent can prevent the cache from working, leading to higher costs. Cache-Assembler fixes these issues automatically.

Benefits

The main benefit of Cache-Assembler is significant cost reduction. By ensuring request prefixes remain identical, it enables prompt-cache hits that can reduce API costs by up to 8.2 times in real-world workloads. The tool offers several key advantages:

1. Deterministic Serialization:It ensures tool definitions are sent with consistent key ordering. This prevents small changes in data structure from breaking the cache.
2. Smart Breakpoint Placement:The tool automatically places cache control breakpoints at high-value boundaries. It uses a tiered strategy with 1-hour and 5-minute time limits without needing manual configuration.
3. Volatile Data Handling:It can move session-unique data out of the stable prefix. This prevents the prefix from changing on every single request.
4. Concurrent Request Coordination:When multiple requests happen at the same time, the tool coordinates them so only one request pays the full cost of writing to the cache. Other requests wait for the first one.
5. Built-in Telemetry:It tracks cache reads and writes, estimates dollar savings, and alerts users when a cached prefix changes unexpectedly.

Use Cases

Cache-Assembler is primarily used by developers who integrate with the Anthropic API. It is especially useful for:

• CLI Tools:It is specifically validated for use with the Claude Code command-line interface. Users can run it as a background proxy service and point their CLI tool to it.
• Web Applications:Developers can import the library directly into their application code to handle request assembly before sending data to the API.
• High-Frequency Workloads:It is ideal for applications that make many requests in a short period, as the concurrent request coordination feature helps share costs effectively.
• Cost-Sensitive Projects:Any project using the Anthropic Messages API can benefit from the potential 8.2x cost reduction.

Pricing

Cache-Assembler is an open-source tool released under the MIT license. It is free to use and install. There are no subscription fees or usage limits for the software itself. The cost savings come from reduced API charges to Anthropic, not from the tool. In a controlled test, the tool reduced costs from $1.33 to $0.16 per 100-turn session.

Vibes

The tool has been validated through a controlled A/B experiment against the real Anthropic API. The results showed an 8.2x reduction in costs. The developers also battle-tested the tool against production traffic and fixed four real bugs related to its features. The community response suggests it is a practical solution for a specific technical problem. Users who rely heavily on the Anthropic API find the cost savings significant enough to justify the setup effort.

Additional Information

Cache-Assembler is built with zero runtime dependencies, making it easy to install and release. The core function is pure and does not mutate input data. The tool follows a fail-open design, meaning if an error occurs during assembly, it logs the issue and forwards the original request instead of breaking the connection. It is available as both a library for code integration and a standalone HTTP proxy. The project includes presets for specific clients like Claude Code to simplify setup.