Zed 1.19.0-pre Update Breakdown: Call Hierarchy, AI Session Optimization, and Git Multi-Select

Overview

Zed editor has released v1.19.0-pre, a preview build of the performance-focused modern code editor. This update touches several core modules including the AI agent, Git panel, and language support. The highlights include Call Hierarchy support, Git panel multi-select, automatic language detection for untitled buffers, and search-on-type enabled by default for project search.

This article walks through the most valuable improvements from a developer's practical perspective, helping you decide whether it's worth upgrading early.

Deep Improvements to the AI Agent

As an editor that's heavily invested in AI-assisted coding, Zed delivered several refinements to AI-related features in this release.

Memory and Streaming Performance

The most immediately useful improvement is reduced memory usage during long Agent sessions (#63053). Memory creep during extended AI conversations or complex tasks has long been a pain point for heavy users — this optimization addresses that directly.

Zed also improved streaming output for OpenAI-compatible providers (#63520) and added better support for Claude Fable 5.1 (#63571). For OpenRouter models, effort-based reasoning support was added (#61308), allowing users to trade off reasoning depth against response speed.

Effort-based reasoning is a parameter control mechanism introduced by OpenAI's o-series models (e.g., o1, o3). It lets users specify low, medium, or high via the reasoning_effort field, balancing the depth of reasoning chains against token consumption and latency. High-effort reasoning suits complex multi-step tasks, while low-effort mode speeds up simple completions or Q&A. OpenRouter, acting as a multi-model aggregation layer, forwards this parameter to supported underlying models via a unified API. Zed's integration means users can control reasoning depth directly from editor settings without manually modifying API request parameters — a real benefit for workflows that frequently switch between speed and quality.

Bug Fixes for Multi-Model Integrations

Several fixes at the model integration layer are also noteworthy: a fix for Gemini rejecting requests when tool schemas couldn't be expressed in Zed's restricted dialect (#63342); a fix for Anthropic's "prompt too long" HTTP 400 errors not being recognized as context window overflow (#63596); and a fix for Ollama not appearing as an edit prediction provider (#63519). These fixes reflect Zed's ongoing effort to polish compatibility across its multi-model ecosystem.

Call Hierarchy: Closing the Gap with Full IDEs

The new feature with the biggest impact on code navigation is Call Hierarchy support (#53239). Developers can now use the call hierarchy: show incoming calls and show outgoing calls commands to see which functions call a given function, and which functions it calls in turn.

The default shortcut is cmd-k cmd-h on macOS, and ctrl-k ctrl-h on Linux and Windows. This is a capability that traditional IDEs like IntelliJ and VS Code have long offered. With Zed now supporting it, the gap between Zed and full-featured IDEs continues to narrow — and it's especially valuable when navigating large codebases.

On the language support front, Shell script recognition was added for .bash_login files (#63466), and preview support (cmd-shift-v) was added for tabular data formats like CSV/TSV (#63369).

Call Hierarchy is a code navigation feature built on the Language Server Protocol (LSP). "Incoming Calls" shows which functions or methods call the current symbol, helping developers quickly locate callers. "Outgoing Calls" shows which other functions are called within the current function, making it easy to understand the dependency chain. Together, these two perspectives let you build a mental model of an unfamiliar codebase's call relationships without reading code line by line. In large monoliths or microservice projects where a single function may be called from dozens of places, Call Hierarchy organizes those references into a tree structure — far more intuitive than a global text search. Zed's implementation relies on each language's LSP server, so actual availability depends on whether the corresponding language server implements the callHierarchy capability (rust-analyzer, clangd, and pyright, among others, already do).

Git Workflow Enhancements

Zed's built-in Git panel has always been a standout feature, and this update continues to strengthen it:

• Git panel multi-select (#60340): You can now stage, discard, and perform other operations on multiple files at once, dramatically improving batch-processing efficiency.
• New Open File on Remote and Copy Remote File URL options (#62177), making it easy to jump directly to the corresponding file in the remote repository.
• Improved performance when opening project diffs with a large number of merge conflicts (#63394, #63395).

Several Git-related bugs were also fixed, including failure to detect remote repositories under partial clones, a worktree naming issue with bare repositories, and the "back" navigation breaking after opening a commit diff from Git Graph.

Editor Experience Polish

Beyond the headline features, Zed continues to refine many details.

Search-on-Type Enabled by Default

This is a breaking change: project search now executes immediately as you type, by default (#62506). Users accustomed to "type first, then press Enter" will need to adjust. To restore the old behavior, add { "search": { "search_on_type": false } } to your settings.

Search-on-type means a full project search is triggered after every keystroke, rather than waiting for the user to press Enter. This feels smooth in small to medium projects, but in large monorepos or workspaces with a massive number of files, every keystroke can trigger a high-load filesystem traversal, causing input lag or CPU spikes. Zed making this the default is a tradeoff decision based on the majority of user scenarios, but it's worth paying close attention to if you work with large codebases. Restoring manual-trigger mode via search_on_type: false ensures searches only run when you press Enter, effectively avoiding performance issues. In semantic versioning terms, this kind of "behavioral default change" counts as a breaking change — confirm whether it suits your workflow before upgrading.

Performance and Persistence Improvements

• Projects using .editorconfig files see improved input and rendering performance (#63323).
• Data persistence is more reliable when Zed exits, restarts after an update, or reloads a workspace (#63213).
• The command palette now sorts results by recency of use, surfacing frequently used commands first (#63388).

New Configuration Options

This release adds several useful settings, including reveal_if_open (reveal a buffer in an already-open panel), folder_indicator (show both the collapse arrow and the folder icon simultaneously), and outline_panel.multi_buffer_hide_symbols (show only files and directories in multi-buffer views). These options further extend the editor's customizability.

Crash and Stability Fixes

As a preview release, this build includes a large number of crash and hang fixes. Notable ones include:

• Fixed a UI hang when language model providers stream large tool inputs (#63241).
• Fixed an Agent terminal tool call leaking file descriptors on macOS (#62978).
• Fixed a crash that could occur when replacing file content in a multi-buffer (#62636).
• Fixed multiple platform-specific crash issues on Windows and Linux.

These fixes demonstrate that Zed's team is maintaining a tight grip on stability even while rapidly shipping new features.

Summary

Zed v1.19.0-pre is an update that balances new features with stability. AI session memory optimization and Call Hierarchy support are the two most practical improvements in this release — the former targets heavy AI users, the latter targets developers working with large codebases. Git multi-select and default search-on-type continue to refine the day-to-day workflow experience.

As a preview build, early adopters can try these features ahead of the stable release, but be mindful of the changed default search behavior. For developers who prioritize performance and native AI integration, Zed's iteration pace remains well worth following.