The post introduces two capabilities on the Claude Developer Platform for managing an agent’s context: Context editing and the memory tool. It starts from a production problem — agents that handle complex tasks and accumulate tool results tend to exhaust their effective context windows, leaving developers to choose between cutting agent transcripts and accepting degraded performance — and presents the two features as complementary answers: one keeps only relevant data in context, the other preserves valuable information across sessions.
Anthropic released both alongside Claude Sonnet 4.5, which it says adds built-in context awareness by tracking available tokens through a conversation. They were announced in public beta on the Claude Developer Platform and through Amazon Bedrock and Google Cloud’s Vertex AI.
Key Points
- Context editing automatically clears stale tool calls and results from the context window when it approaches token limits, preserving the conversation flow; the post says this extends how long agents can run without manual intervention and improves effective model performance because Claude focuses only on relevant context.
- The memory tool lets Claude store and consult information outside the context window through a file-based system: Claude can create, read, update and delete files in a dedicated memory directory that persists across conversations (see Structured note-taking).
- The memory tool operates entirely client-side through tool calls, and the storage backend is in the developer’s own infrastructure, so developers control where and how the data is persisted.
- The post’s suggested division of labor by use case: in coding, context editing clears old file reads and test results while memory keeps debugging insights and architectural decisions; in research, memory stores key findings while old search results are cleared; in data processing, intermediate results go to memory while raw data is cleared.
- On an internal evaluation set for agentic search, Anthropic reports that combining the memory tool with context editing improved performance by 39% over baseline, and context editing alone by 29%.
- In a 100-turn web search evaluation, it reports that context editing let agents complete workflows that would otherwise have failed from context exhaustion, while cutting token consumption by 84%.
Context
This is a vendor’s product announcement, and its performance figures come from Anthropic’s own internal evaluations. The two capabilities it introduces correspond to techniques this wiki covers in more depth elsewhere: clearing tool results from the transcript (Tool-result clearing) and agentic memory (Structured note-taking), both compared with Compaction in Context engineering: memory, compaction, and tool clearing.