Zed 1.17.2 Released: Table Preview, AI Agent Tools, and Large File Performance Improvements

Zed 1.17.2 adds table preview, AI ask_user tool, large file memory cuts, and cross-platform fixes.
Zed 1.17.2 advances on three fronts: table data preview for CSV/TSV and similar formats is now available to all users with sorting, filtering, and column resizing; the AI Agent layer gains Gemini 2.5 Flash support and the new `ask_user` tool for clarifying requirements mid-task (disabled by default to avoid workflow interruptions); and large file memory usage drops by roughly one file-size worth of peak allocation, alongside rendering speed improvements and several resource leak fixes. Git integration adds blame history navigation and tiered stash operations, while cross-platform fixes address CRLF handling and Windows remote server path construction.
The high-performance code editor Zed has landed its 1.17.2 update. This release focuses on three dimensions: data readability, AI Agent interaction, and large file performance — along with a range of improvements to Git, Vim/Helix modes, and cross-platform compatibility. As a modern editor built in Rust and optimized for raw speed, every Zed iteration is worth a developer's attention.
Table Data Preview: Structured Data Browsing for All Users
One of the most eye-catching features in this update is table data preview, now available to all users. Zed natively supports visual browsing of CSV, TSV, PSV, and SSV files with a surprisingly complete feature set:
- Columns support sorting and can be resized independently;
- Row filtering based on selected values across one or more columns;
- A persistent row number column that stays visible during horizontal scrolling;
- Right-click support for copying cell and column header content.
This feature is the result of more than a dozen PRs (#48207, #53295, #62773, and others), led by community contributor HalavicH. For developers and data engineers who regularly work with data files, the ability to review structured data directly in the editor — without switching to a dedicated spreadsheet tool — offers a meaningful productivity boost.

AI Agent Upgrades: New Models and Human-in-the-Loop Tooling
AI-assisted programming has been a key focus of Zed's recent development, and this release advances both model support and interaction patterns.
New Models and Subscription Experience Improvements
Zed adds support for Google's Gemini 2.5 Flash model (#62670), and introduces a low reasoning effort option for DeepSeek V3 and V3 Pro (#62577). On the subscription side, the team improved ChatGPT subscription caching and optimized reasoning continuity and request privacy for OpenRouter (#62652).
The ask_user Tool: A More Collaborative Human-AI Paradigm
Perhaps the most notable addition is the new ask_user tool (#61497). It allows an Agent to prompt users for input via a form, supporting selectable options, free-text input, or a combination of both. This means the AI can actively clarify requirements before proceeding — rather than forging ahead blindly — representing a more mature human-AI collaboration model.
That said, Zed is being notably restrained here: in this release's breaking changes, the team explicitly disables ask_user by default (#63038). This decision reflects a deliberate caution around Agents interrupting user workflows — the capability is provided, but users must opt in.
The Agent sidebar also gains move project up/down actions for reordering projects (#62695), continuing the team's attention to polish.
Performance: Lower Memory Usage and Faster Rendering for Large Files
Performance has always been Zed's core value proposition, and this update delivers meaningful improvements across several areas.
The most impactful change is a significant reduction in memory usage when opening large files — peak memory during loading is reduced by approximately the size of the file itself (#62748). For log files or data files in the hundreds of megabytes, this is a substantial improvement. Editor rendering speed also received a general boost (#62715).
Several resource leak issues were also addressed:
- A memory leak caused by background scans continuing after worktree creation was cancelled, leading to unbounded snapshot growth (#61009);
- Excessive CPU and memory usage when opening large directory trees not tracked by Git (#62583);
- A performance bottleneck in project search when
file_scan_inclusionswas overly broad (#62658).
To help developers diagnose performance issues on their own, Zed adds a frame time debug overlay that can be cycled through with ctrl-alt-shift-p (#62749).
Git and Editor Workflow Enhancements
This release fills in several commonly needed Git operations:
- New
editor: blame revisionandeditor: blame previous revisioncommands for viewing blame at a specific commit or its parent (#62614); - The Git panel gains
Stash TrackedandStash Stagedoptions, allowing only tracked or staged changes to be stashed (#62254); - Collapse/expand behavior in the Git panel has been aligned with the project panel for consistency (#61846).
In Vim/Helix mode, the h and l keys can now collapse and expand entries in the Git panel (#61846), smoothing keyboard-driven workflows. JetBrains keymap support gains CamelHump-style subword navigation (#51540).
Stability and Cross-Platform Compatibility Fixes
This release includes a large number of bug fixes, many addressing cross-platform consistency and correct LSP behavior:
- Fixed incorrect project path construction when connecting to a Windows remote server from a Unix machine (#62038);
- Fixed language servers receiving incorrect line endings for CRLF files, causing false positives from linters and formatters (#59941);
- Fixed Zed update downloads stalling after the system wakes from sleep (#60366);
- Fixed manual indentation being lost when using an input method or replacing inline content (#62644).
macOS users also gain a simple option for fullscreen_mode to handle the case where fullscreen windows overlap the MacBook notch (#60020).
Summary
Zed 1.17.2 is a classic "inside and out" release: outward-facing improvements to productivity through table preview and AI Agent capabilities, while the internals continue to be refined around memory usage, rendering performance, and cross-platform compatibility. The deliberate choice to disable ask_user by default, and the practical large-file memory optimization, both reflect a team that pursues speed without compromising engineering quality. For developers looking for a lightweight, high-performance, AI-native code editor, Zed remains a compelling option worth following.
Related articles

Catalyst: A Vision for an Enzyme-Like Testing Framework for AI Agents
A developer shared Catalyst on Reddit, an Enzyme-inspired framework for AI Agents, exploring why agents need observable, testable dev tools and the design philosophy behind them.

The Real Capability of AI Coding Agents: Best Models Complete Only 35% of Feature Development Tasks
The 'Agents on Rails' benchmark finds top AI models complete only 35% of feature development tasks. What this means for coding agents and developer teams.

How to Prevent Duplicate Refunds After an AI Agent Crashes: CellaFlow's Durable Execution Approach
How can AI agents avoid duplicate refunds after a crash without deadlocking workflows? CellaFlow uses durable execution, shared work identity, leases, and fencing to solve safety and liveness in multi-agent systems.