zed v1.25.2
- Source
- github.com
- Author
- zed-zippy[bot]
- Date

Context
The change in Zed v1.25.2 most worth knowing about is a security fix for remote development. When a project was open on a remote machine, environment variables that Zed passed to external agents and servers, "including secrets," were visible in the local machine's process list. Environment variables are a common way to hand API keys and tokens to those tools, and a process list is something other programs on the computer can usually read, as can other users on a shared machine. The release notes say only that this is fixed; they don't say how the values ended up there, which earlier versions were affected, or whether anyone should rotate keys, and the linked pull request could not be read for this explanation.
The release also gives external agents two new features. In Zed's terms, external agents are tools such as Claude , Codex or Gemini CLI that run as separate processes and talk to the editor over the Agent Client Protocol, ACP, while usually keeping their own models, sign-in and tools. They now get session notices and . Compaction is the technique of summarizing the older part of a long conversation so it still fits in the model's , letting a long session continue instead of hitting the limit. The notes give this a single line and don't describe how Zed and each agent divide that work. A related fix addresses prompts to external agents that were slow to send, or never sent, because of unneeded Git checkpoints.
Two smaller items are worth a sentence. Zed's binary is about 33% smaller on macOS, 28% on Windows and 7% on Linux, and the Amazon Bedrock provider adds GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5.
- Zed v1.25.2 fixes a bug where environment variables, including secrets, passed to external agents and MCP servers in remote projects were visible in the local machine's process list.
- Prompts to external agents could be slow to send or never send because Zed created unneeded Git checkpoints; that is fixed, as is the occasional loss of an external agent's first updates in a new thread.
- Zed's binaries shrink by about 33% on macOS, 28% on Windows and 7% on Linux, and SSH remote connections now detect shell, platform and OS version in one command instead of three, which helps most when connecting from Windows.
- Gemini models no longer fail after a tool call returns an image, and OpenAI-compatible edit predictions now work on servers that limit stop sequences to 4.
- MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context compaction — Summarizing an agent's earlier conversation to free room in the context window so a long session can keep going.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…


