Managing context on the Claude Developer Platform (www.anthropic.com)

🤖 AI Summary
Anthropic today launched context editing and a client-side memory tool on the Claude Developer Platform alongside Claude Sonnet 4.5, letting agents manage and persist context for much longer, more complex workflows. Context editing automatically prunes stale tool calls and results from the active context as token limits approach, while the memory tool exposes a file-based memory directory (create/read/update/delete via tool calls) that developers host and control. Sonnet 4.5 adds built-in context-awareness by tracking available tokens, so agents can prioritize relevant information and avoid losing critical state across sessions. Technically, the combination addresses the classic context-window bottleneck for agentic systems: editing reduces irrelevant tokens in-flight, and memory offloads persistent knowledge outside the window while remaining queryable. Anthropic reports large gains on internal agentic benchmarks—context editing alone improved performance 29%, and combined with the memory tool delivered a 39% boost; a 100-turn web search test showed workflows that would have failed for context exhaustion completed successfully while cutting token consumption by 84%. The features are in public beta (also available via Amazon Bedrock and Google Vertex AI) and target long-running use cases like multi-file coding, research knowledge bases, and large-scale data processing where maintaining project state and tool histories is essential.
Loading comments...
loading comments...